feat\!: rename baoyu-imagine→baoyu-image-gen, baoyu-image-cards→baoyu-xhs-images (v2.0.0)

BREAKING CHANGE: removed `baoyu-imagine` and `baoyu-image-cards`. All functionality now lives under `baoyu-image-gen` and `baoyu-xhs-images` respectively. Cross-skill `## Image Generation Tools` examples updated across baoyu-article-illustrator, baoyu-comic, baoyu-cover-image, baoyu-infographic, and baoyu-slide-deck. Migration: existing `~/.baoyu-skills/baoyu-imagine/EXTEND.md` configs are auto-renamed to `…/baoyu-image-gen/EXTEND.md` on first run via the legacy-path resolver in `scripts/main.ts`.
2026-08-08 01:43:03 +08:00 · 2026-05-24 18:35:05 -05:00
parent bc303345f4
commit c3cbce9ce3
116 changed files with 2117 additions and 12789 deletions
@@ -1,7 +1,7 @@
 ---
 name: baoyu-image-gen
-description: "[Deprecated: use baoyu-imagine] AI image generation with OpenAI, Azure OpenAI, Google, OpenRouter, DashScope, Z.AI GLM-Image, MiniMax, Jimeng, Seedream and Replicate APIs. Supports text-to-image, reference images, aspect ratios, and batch generation from saved prompt files. Sequential by default; use batch parallel generation when the user already has multiple prompts or wants stable multi-image throughput. Use when user asks to generate, create, or draw images."
-version: 1.56.4
+description: AI image generation with OpenAI GPT Image 2, Azure OpenAI, Google, OpenRouter, DashScope, Z.AI GLM-Image, MiniMax, Jimeng, Seedream and Replicate APIs. Supports text-to-image, reference images, aspect ratios, and batch generation from saved prompt files. Sequential by default; use batch parallel generation when the user already has multiple prompts or wants stable multi-image throughput. Use when user asks to generate, create, or draw images.
+version: 2.0.0
 metadata:
  openclaw:
    homepage: https://github.com/JimLiu/baoyu-skills#baoyu-image-gen
@@ -11,10 +11,9 @@ metadata:
        - npx
 ---

-
 # Image Generation (AI SDK)

-Official API-based image generation. Supports OpenAI, Azure OpenAI, Google, OpenRouter, DashScope (阿里通义万象), Z.AI GLM-Image, MiniMax, Jimeng (即梦), Seedream (豆包) and Replicate.
+Official API-based image generation. Supports OpenAI GPT Image 2, Azure OpenAI, Google, OpenRouter, DashScope (阿里通义万象), Z.AI GLM-Image, MiniMax, Jimeng (即梦), Seedream (豆包) and Replicate.

 ## User Input Tools

@@ -45,6 +44,8 @@ Check these paths in order; first hit wins:
 - **Found** → load, parse, apply. If `default_model.[provider]` is null → ask model only.
 - **Not found** → run first-time setup (`references/config/first-time-setup.md`) using AskUserQuestion to collect provider + model + quality + save location. Save EXTEND.md, then continue. Do not generate images before this completes.

+Legacy compatibility: if `.baoyu-skills/baoyu-imagine/EXTEND.md` exists and the new path doesn't, the runtime renames it to `baoyu-image-gen`. If both exist, the runtime leaves them alone and uses the new path.
+
 **EXTEND.md keys**: default provider, default quality, default aspect ratio, default image size, OpenAI image API dialect, default models, batch worker cap, provider-specific batch limits. Schema: `references/config/preferences-schema.md`.

 ## Usage
@@ -77,6 +78,9 @@ ${BUN_X} {baseDir}/scripts/main.ts --prompt "Make blue" --image out.png --ref so
 # Specific provider
 ${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider dashscope --model qwen-image-2.0-pro

+# OpenAI GPT Image 2
+${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider openai --model gpt-image-2
+
 # Batch mode
 ${BUN_X} {baseDir}/scripts/main.ts --batchfile batch.json --jobs 4
 ```
@@ -103,11 +107,11 @@ When the user wants a person/object preserved from reference images:
 | `--provider google\|openai\|azure\|openrouter\|dashscope\|zai\|minimax\|jimeng\|seedream\|replicate` | Force provider (default: auto-detect) |
 | `--model <id>`, `-m` | Model ID — see provider references for defaults and allowed values |
 | `--ar <ratio>` | Aspect ratio (`16:9`, `1:1`, `4:3`, …) |
-| `--size <WxH>` | Explicit size (e.g., `1024x1024`) |
+| `--size <WxH>` | Explicit size (e.g., `1024x1024`; for `gpt-image-2`, width/height must be multiples of 16, max edge 3840px, ratio no wider than 3:1) |
 | `--quality normal\|2k` | Quality preset (default: `2k`) |
 | `--imageSize 1K\|2K\|4K` | Image size for Google/OpenRouter (default: from quality) |
 | `--imageApiDialect openai-native\|ratio-metadata` | OpenAI-compatible endpoint dialect — use `ratio-metadata` for gateways that expect aspect-ratio `size` plus `metadata.resolution` |
-| `--ref <files...>` | Reference images. Supported by Google multimodal, OpenAI GPT Image edits, Azure OpenAI edits (PNG/JPG only), OpenRouter multimodal models, Replicate supported families, MiniMax subject-reference, Seedream 5.0/4.5/4.0. Not supported by Jimeng, Seedream 3.0, SeedEdit 3.0 |
+| `--ref <files...>` | Reference images. Supported by Google multimodal, OpenAI GPT Image edits, Azure OpenAI edits (PNG/JPG only), OpenRouter multimodal models, Replicate supported families, MiniMax subject-reference, Seedream 5.0/4.5/4.0, DashScope `wan2.7-image-pro`/`wan2.7-image`. Not supported by Jimeng, Seedream 3.0, SeedEdit 3.0, or any DashScope model outside the `wan2.7-image*` family |
 | `--n <count>` | Number of images. Replicate requires `--n 1` (single-output save semantics) |
 | `--json` | JSON output |

@@ -140,15 +144,15 @@ When the user wants a person/object preserved from reference images:

 ### Codex/ChatGPT OAuth is not an OpenAI API key

-`--provider openai` uses the standard OpenAI Images API and requires `OPENAI_API_KEY`. A Codex or ChatGPT desktop login is a different entitlement and is not a drop-in replacement for `OPENAI_API_KEY`; do not paste a Codex OAuth token into `OPENAI_API_KEY` or only set `OPENAI_BASE_URL` to a Codex backend.
+`--provider openai --model gpt-image-2` uses the standard OpenAI Images API (`/v1/images/generations` or `/v1/images/edits`) and requires `OPENAI_API_KEY`. A Codex or ChatGPT desktop login is a different entitlement and is not a drop-in replacement for `OPENAI_API_KEY`; do not paste a Codex OAuth token into `OPENAI_API_KEY` or only set `OPENAI_BASE_URL` to a Codex backend.

 If the user wants to use their Codex subscription / GPT Image 2 entitlement without an OpenAI API key, route through a Codex-native backend instead of this skill's `openai` provider:

 - In Codex runtime: use the native `imagegen` skill/tool.
- In non-Codex runtimes with `codex` CLI installed and logged in: use the repo-level `scripts/codex-imagegen.sh` wrapper when the calling skill supports it. Resolve it from the plugin/repo root and pass absolute prompt/output/reference paths.
+- In non-Codex runtimes with `codex` CLI installed and logged in: use the repo-level `scripts/codex-imagegen.sh` wrapper when the calling skill supports it (for example `baoyu-cover-image`). Resolve it from the plugin/repo root and pass absolute prompt/output/reference paths.
 - In Hermes runtimes with a native `image_generate` tool: use that tool as a fallback, and state whether reference images were passed directly or reconstructed from extracted traits.

-Do not modify the existing `openai` provider to silently consume Codex OAuth. If first-class Codex OAuth support is added, implement it as a distinct provider (for example `openai-codex`) with its own auth, route, request shape, docs, and tests. See `references/codex-oauth-vs-openai-api-key.md`.
+Do not modify the existing `openai` provider to silently consume Codex OAuth. If first-class Codex OAuth support is added to `baoyu-image-gen`, implement it as a distinct provider (for example `openai-codex`) with its own auth, route, request shape, docs, and tests. See `references/codex-oauth-vs-openai-api-key.md`.

 ## Model Resolution

@@ -159,7 +163,9 @@ Priority (highest → lowest) applies to every provider:
 3. Env var `<PROVIDER>_IMAGE_MODEL`
 4. Built-in default

-For Azure, `--model` / `default_model.azure` is the Azure deployment name. `AZURE_OPENAI_DEPLOYMENT` is the preferred env var; `AZURE_OPENAI_IMAGE_MODEL` is kept as a backward-compatible alias.
+For OpenAI, the built-in default is `gpt-image-2`. `gpt-image-1.5`, `gpt-image-1`, and GPT Image snapshots remain selectable with `--model` or `OPENAI_IMAGE_MODEL`.
+
+For Azure, `--model` / `default_model.azure` is the Azure deployment name. `AZURE_OPENAI_DEPLOYMENT` is the preferred env var; `AZURE_OPENAI_IMAGE_MODEL` is kept as a backward-compatible alias. If your Azure deployment is named after the underlying model, use `gpt-image-2`; otherwise use the exact custom deployment name.

 EXTEND.md overrides env vars: if EXTEND.md sets `default_model.google: "gemini-3-pro-image-preview"` and the env var sets `GOOGLE_IMAGE_MODEL=gemini-3.1-flash-image-preview`, EXTEND.md wins.

@@ -200,17 +206,19 @@ Each provider has its own quirks (model families, size rules, ref support, limit

 | Preset | Google imageSize | OpenAI size | OpenRouter size | Replicate resolution | Use case |
 |--------|------------------|-------------|-----------------|----------------------|----------|
-| `normal` | 1K | 1024px | 1K | 1K | Quick previews |
-| `2k` (default) | 2K | 2048px | 2K | 2K | Covers, illustrations, infographics |
+| `normal` | 1K | 1024px target | 1K | 1K | Quick previews |
+| `2k` (default) | 2K | 2048px target | 2K | 2K | Covers, illustrations, infographics |

 Google/OpenRouter `imageSize` can be overridden with `--imageSize 1K|2K|4K`.

+For OpenAI native `gpt-image-2`, `normal` maps to `quality=medium` and a low-latency valid size near the requested aspect ratio; `2k` maps to `quality=high` and 2048px-class sizes such as `2048x2048`, `2048x1152`, or `1152x2048`. Use explicit `--size` for valid custom or 4K outputs, e.g. `3840x2160`.
+
 ## Aspect Ratios

 Supported: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `2.35:1`.

 - Google multimodal: `imageConfig.aspectRatio`
- OpenAI: closest supported size
+- OpenAI: `gpt-image-2` uses the closest valid custom size for the requested ratio; older GPT Image and DALL·E models use their closest supported fixed size
 - OpenRouter: `imageGenerationOptions.aspect_ratio`; if only `--size <WxH>` is given, the ratio is inferred
 - Replicate: behavior is model-specific — `google/nano-banana*` uses `aspect_ratio`, `bytedance/seedream-*` uses documented Replicate ratios, Wan 2.7 maps `--ar` to a concrete `size`
 - MiniMax: official `aspect_ratio` values; if `--size <WxH>` is given without `--ar`, sends `width`/`height` for `image-01`
@@ -245,14 +253,14 @@ Rule of thumb: once prompt files are saved and the task is "generate all of thes

 ### Codex image2 fallback

-If `--provider openai` fails because `OPENAI_API_KEY` is missing but the current runtime has a native image-generation backend or the repo-level `codex-imagegen` wrapper is available, use that path rather than leaving the user waiting. Be explicit about whether the fallback is true reference-image generation or only a text-prompt reconstruction from extracted visual traits. See `references/codex-image2-fallback.md`.
+If `--provider openai --model gpt-image-2` fails because `OPENAI_API_KEY` is missing but the current runtime has a native image-generation backend or the repo-level `codex-imagegen` wrapper is available, use that path rather than leaving the user waiting. Be explicit about whether the fallback is true reference-image generation or only a text-prompt reconstruction from extracted visual traits. See `references/codex-image2-fallback.md`.

 ## References

 | File | Content |
 |------|---------|
 | `references/usage-examples.md` | Extended CLI examples across providers and batch mode |
-| `references/codex-oauth-vs-openai-api-key.md` | Why Codex/ChatGPT OAuth image2 entitlement is not usable through the standard OpenAI API-key provider |
+| `references/codex-oauth-vs-openai-api-key.md` | Why Codex/ChatGPT OAuth image2 entitlement is not usable through baoyu-image-gen's standard OpenAI API-key provider |
 | `references/codex-image2-fallback.md` | Practical fallback behavior when OpenAI API credentials are absent but Codex/native image generation is available |
 | `references/providers/dashscope.md` | DashScope families, sizes, limits |
 | `references/providers/zai.md` | Z.AI GLM-image / cogview-4 |
@@ -1,11 +1,11 @@
 ---
 name: codex-image2-fallback
-description: Fallback behavior when OpenAI API credentials are absent but Codex/native image generation is available
+description: Fallback behavior when baoyu-image-gen lacks OpenAI API credentials but Codex/native image generation is available
 ---

 # Codex Image2 Fallback

-The `openai` provider can fail with:
+When using `baoyu-image-gen` with `--provider openai --model gpt-image-2`, the CLI can fail with:

 ```text
 OPENAI_API_KEY is required. Codex/ChatGPT desktop login does not automatically grant OpenAI Images API access to this script.
@@ -15,7 +15,7 @@ This is expected. The `openai` provider uses the public OpenAI Images API and ne

 ## Practical fallback pattern

-1. Try this skill when provider credentials are available.
+1. Try `baoyu-image-gen` when provider credentials are available.
 2. If it fails only because `OPENAI_API_KEY` is missing, do not leave the user waiting.
 3. Prefer a Codex/native raster backend in this order:
   - Codex runtime native `imagegen` skill/tool, if available.
@@ -25,3 +25,11 @@ This is expected. The `openai` provider uses the public OpenAI Images API and ne
   - If the fallback backend accepts references, pass the reference images.
   - If it does not, derive a concise identity-preserving prompt from the references and state that it is a text-description fallback, not strict reference-image editing.
 5. Return the generated media path or structured backend error promptly.
+
+## User-facing wording
+
+Use concise wording such as:
+
+> The OpenAI API path needs `OPENAI_API_KEY`; Codex login is a separate image2 backend. I used the available Codex/native image backend instead. Reference images were [passed directly / reconstructed from visual traits].
+
+Avoid implying that `baoyu-image-gen --provider openai` can use Codex OAuth without a dedicated provider implementation.
@@ -1,20 +1,18 @@
-# Codex OAuth vs OpenAI API key
+# Codex OAuth vs OpenAI API key for baoyu-image-gen

-This deprecated skill keeps cross-cutting image-generation guidance in sync with `baoyu-imagine`.
+`baoyu-image-gen --provider openai` uses the standard OpenAI Images API and requires `OPENAI_API_KEY`. It calls OpenAI-compatible image endpoints such as `/images/generations` and `/images/edits`.

-The `openai` provider uses the standard OpenAI Images API and requires `OPENAI_API_KEY`. It calls OpenAI-compatible image endpoints such as `/images/generations` and `/images/edits`.
-
-Codex / ChatGPT login is different. Codex image generation is driven by Codex OAuth and the Codex runtime's `image_gen` capability, not by the public OpenAI Images API key path. A Codex OAuth token is not a drop-in replacement for `OPENAI_API_KEY`, and setting `OPENAI_BASE_URL` to a Codex backend will not make the existing `openai` provider work because the auth, route, and payload shape differ.
+Codex / ChatGPT login is different. Codex image generation is driven by Codex OAuth and the Codex runtime's `image_gen` capability, not by the public OpenAI Images API key path. A Codex OAuth token is not a drop-in replacement for `OPENAI_API_KEY`, and setting `OPENAI_BASE_URL` to a Codex backend will not make baoyu-image-gen's existing `openai` provider work because the auth, route, and payload shape differ.

 ## What to use instead

 - If running inside Codex and the native `imagegen` skill/tool is available, use it directly.
- If running outside Codex but the `codex` CLI is installed and logged in, use the repo-level `scripts/codex-imagegen.sh` wrapper when the calling skill supports it.
+- If running outside Codex but the `codex` CLI is installed and logged in, use the repo-level `scripts/codex-imagegen.sh` wrapper when the calling skill supports it. The wrapper invokes `codex exec` and the Codex `image_gen` tool; no `OPENAI_API_KEY` is required.
 - If running inside Hermes and a native `image_generate` tool is available, use that as a runtime-native fallback. Be explicit about whether reference images are passed directly or only reconstructed from extracted traits.
- If first-class Codex OAuth support is added, add a distinct provider such as `openai-codex` rather than modifying the existing `openai` provider.
+- If the user wants `baoyu-image-gen` itself to support Codex OAuth, add a distinct provider such as `openai-codex` rather than modifying the existing `openai` provider.

 ## Reference-image prompting note

-When using actual reference images for identity preservation, avoid long generic descriptions of the subject. Prefer direct wording:
+When using actual reference images for identity preservation, avoid long generic descriptions of the subject. Long descriptions can cause the model to synthesize a new similar-looking person/object. Prefer direct wording:

 > Use the person/object in the reference image(s) as the same identity. Do not redesign it or create a similar-looking new subject. Only change scene, clothing, pose, lighting, rendering style, and composition.
@@ -46,19 +46,19 @@ options:
  - label: "Google (Recommended)"
    description: "Gemini multimodal - high quality, reference images, flexible sizes"
  - label: "OpenAI"
-    description: "GPT Image - consistent quality, reliable output"
+    description: "GPT Image 2 - latest OpenAI image model, reference-image workflows"
  - label: "Azure OpenAI"
    description: "Azure-hosted GPT Image deployments with resource-specific routing"
  - label: "OpenRouter"
    description: "Router for Gemini/FLUX/OpenAI-compatible image models"
  - label: "DashScope"
    description: "Alibaba Cloud - Qwen-Image, strong Chinese/English text rendering"
+  - label: "Z.AI"
+    description: "GLM-image, strong poster and text-heavy image generation"
  - label: "MiniMax"
    description: "MiniMax image generation with subject-reference character workflows"
  - label: "Replicate"
-    description: "Community models - nano-banana-pro, flexible model selection"
-  - label: "Z.AI"
-    description: "GLM-Image - text-to-image with recommended aspect sizes"
+    description: "Curated Replicate image families - nano-banana-2, Seedream, and Wan image models"
 ```

 ### Question 2: Default Google Model
@@ -101,10 +101,12 @@ Only show if user selected Azure OpenAI.
 header: "Azure Deploy"
 question: "Default Azure image deployment name?"
 options:
-  - label: "gpt-image-1.5 (Recommended)"
-    description: "Best default if your Azure deployment uses the same name"
-  - label: "gpt-image-1"
+  - label: "gpt-image-2 (Recommended)"
+    description: "Use if your Azure deployment uses the GPT Image 2 model name"
+  - label: "gpt-image-1.5"
    description: "Previous GPT Image deployment name"
+  - label: "gpt-image-1"
+    description: "Earlier GPT Image deployment name"
 ```

 ### Question 2d: Default MiniMax Model
@@ -130,11 +132,9 @@ header: "Z.AI Model"
 question: "Default Z.AI image generation model?"
 options:
  - label: "glm-image (Recommended)"
-    description: "Latest GLM-Image, best aspect-ratio coverage and text rendering"
+    description: "Best default for posters, diagrams, and text-heavy images"
  - label: "cogview-4-250304"
-    description: "Legacy CogView-4 model with 16-pixel size stepping"
-  - label: "cogview-4"
-    description: "Previous CogView-4 snapshot for compatibility"
+    description: "Legacy Z.AI image model on the same endpoint"
 ```

 ### Question 3: Default Quality
@@ -177,18 +177,21 @@ default_provider: [selected provider or null]
 default_quality: [selected quality]
 default_aspect_ratio: null
 default_image_size: null
+default_image_api_dialect: null
 default_model:
  google: [selected google model or null]
  openai: null
  azure: [selected azure deployment or null]
  openrouter: [selected openrouter model or null]
  dashscope: null
+  zai: [selected Z.AI model or null]
  minimax: [selected minimax model or null]
  replicate: null
-  zai: [selected zai model or null]
 ---
 ```

+If the user selects `OpenAI` but says their endpoint is only OpenAI-compatible and fronts another image model family, save `default_image_api_dialect: ratio-metadata` when they explicitly confirm the gateway expects aspect-ratio `size` plus metadata-based resolution. Otherwise leave it `null` / `openai-native`.
+
 ## Flow 2: EXTEND.md Exists, Model Null

 When EXTEND.md exists but `default_model.[current_provider]` is null, ask ONLY the model question for the current provider.
@@ -213,10 +216,12 @@ options:
 header: "OpenAI Model"
 question: "Choose a default OpenAI image generation model?"
 options:
-  - label: "gpt-image-1.5 (Recommended)"
-    description: "Latest GPT Image model, high quality"
+  - label: "gpt-image-2 (Recommended)"
+    description: "Latest GPT Image model, flexible sizes up to 4K, high-fidelity image inputs"
+  - label: "gpt-image-1.5"
+    description: "Previous GPT Image model"
  - label: "gpt-image-1"
-    description: "Previous generation GPT Image model"
+    description: "Earlier GPT Image model"
 ```

 ### Azure Deployment Selection
@@ -225,8 +230,10 @@ options:
 header: "Azure Deploy"
 question: "Choose a default Azure image deployment name?"
 options:
-  - label: "gpt-image-1.5 (Recommended)"
-    description: "Use when your Azure deployment name matches the GPT-image-1.5 model"
+  - label: "gpt-image-2 (Recommended)"
+    description: "Use when your Azure deployment name matches the GPT Image 2 model"
+  - label: "gpt-image-1.5"
+    description: "Use when your Azure deployment name matches the GPT Image 1.5 model"
  - label: "gpt-image-1"
    description: "Use when your Azure deployment name matches GPT-image-1"
 ```
@@ -264,6 +271,10 @@ options:
    description: "Legacy Qwen model with five fixed output sizes"
  - label: "qwen-image-plus"
    description: "Legacy Qwen model, same current capability as qwen-image"
+  - label: "wan2.7-image-pro"
+    description: "Wan 2.7 Pro — supports up to 4K text-to-image and reference-image editing"
+  - label: "wan2.7-image"
+    description: "Wan 2.7 base — faster generation, up to 2K, supports reference-image editing"
  - label: "z-image-turbo"
    description: "Legacy DashScope model for compatibility"
  - label: "z-image-ultra"
@@ -274,18 +285,41 @@ Notes for DashScope setup:

 - Prefer `qwen-image-2.0-pro` when the user needs custom `--size`, uncommon ratios like `21:9`, or strong Chinese/English text rendering.
 - `qwen-image-max` / `qwen-image-plus` / `qwen-image` only support five fixed sizes: `1664*928`, `1472*1104`, `1328*1328`, `1104*1472`, `928*1664`.
+- `wan2.7-image-pro` and `wan2.7-image` are the only DashScope models that accept `--ref`. Pick one of these when the user wants reference-image editing or multi-image fusion via DashScope.
 - In `baoyu-image-gen`, `quality` is a compatibility preset. It is not a native DashScope parameter.

+### Z.AI Model Selection
+
+```yaml
+header: "Z.AI Model"
+question: "Choose a default Z.AI image generation model?"
+options:
+  - label: "glm-image (Recommended)"
+    description: "Current flagship image model with better text rendering and poster layouts"
+  - label: "cogview-4-250304"
+    description: "Legacy model on the sync image endpoint"
+```
+
+Notes for Z.AI setup:
+
+- Prefer `glm-image` for posters, diagrams, and Chinese/English text-heavy layouts.
+- In `baoyu-image-gen`, Z.AI currently exposes text-to-image only; reference images are not wired for this provider.
+- The sync Z.AI image API returns a downloadable image URL, which the runtime saves locally after download.
+
 ### Replicate Model Selection

 ```yaml
 header: "Replicate Model"
 question: "Choose a default Replicate image generation model?"
 options:
-  - label: "google/nano-banana-pro (Recommended)"
-    description: "Google's fast image model on Replicate"
-  - label: "google/nano-banana"
-    description: "Google's base image model on Replicate"
+  - label: "google/nano-banana-2 (Recommended)"
+    description: "Current default for general Replicate image generation in baoyu-image-gen"
+  - label: "bytedance/seedream-4.5"
+    description: "Replicate Seedream 4.5 with validated local size/ref guardrails"
+  - label: "bytedance/seedream-5-lite"
+    description: "Replicate Seedream 5 Lite with validated local size/ref guardrails"
+  - label: "wan-video/wan-2.7-image-pro"
+    description: "Replicate Wan 2.7 Image Pro with 4K text-to-image support"
 ```

 ### MiniMax Model Selection
@@ -306,27 +340,6 @@ Notes for MiniMax setup:
 - `image-01-live` is useful when the user prefers faster generation and can work with aspect-ratio-based sizing.
 - MiniMax subject reference currently uses `subject_reference[].type = character`; docs recommend front-facing portrait references in JPG/JPEG/PNG under 10MB.

-### Z.AI Model Selection
-
-```yaml
-header: "Z.AI Model"
-question: "Choose a default Z.AI image generation model?"
-options:
-  - label: "glm-image (Recommended)"
-    description: "Latest GLM-Image; pixels round to multiples of 32 and cap at 2^22"
-  - label: "cogview-4-250304"
-    description: "Legacy CogView-4 snapshot with 16-pixel size stepping"
-  - label: "cogview-4"
-    description: "Earlier CogView-4 snapshot for compatibility"
-```
-
-Notes for Z.AI setup:
-
- Set `ZAI_API_KEY` (or legacy `BIGMODEL_API_KEY`) from https://docs.z.ai/.
- `glm-image` supports recommended aspect sizes (1280x1280, 1728x960, 1568x1056, …); uncommon ratios auto-fit to the 2^22 pixel budget on multiples of 32.
- Legacy CogView models use 16-pixel stepping and cap at 2^21 pixels per image.
- Z.AI does not accept reference images or `n > 1` in `baoyu-image-gen`; use Google/OpenAI providers for those workflows.
-
 ### Update EXTEND.md

 After user selects a model:
@@ -342,9 +355,9 @@ default_model:
  azure: [value or null]
  openrouter: [value or null]
  dashscope: [value or null]
+  zai: [value or null]
  minimax: [value or null]
  replicate: [value or null]
-  zai: [value or null]
 ```

 Only set the selected provider's model; leave others as their current value or null.
@@ -11,7 +11,7 @@ description: EXTEND.md YAML schema for baoyu-image-gen user preferences
 ---
 version: 1

-default_provider: null      # google|openai|azure|openrouter|dashscope|minimax|replicate|zai|null (null = auto-detect)
+default_provider: null      # google|openai|azure|openrouter|dashscope|zai|minimax|replicate|null (null = auto-detect)

 default_quality: null       # normal|2k|null (null = use default: 2k)

@@ -19,15 +19,17 @@ default_aspect_ratio: null  # "16:9"|"1:1"|"4:3"|"3:4"|"2.35:1"|null

 default_image_size: null    # 1K|2K|4K|null (Google/OpenRouter, overrides quality)

+default_image_api_dialect: null  # openai-native|ratio-metadata|null (OpenAI-compatible gateways; null = use env/default)
+
 default_model:
  google: null              # e.g., "gemini-3-pro-image-preview", "gemini-3.1-flash-image-preview"
-  openai: null              # e.g., "gpt-image-1.5", "gpt-image-1"
-  azure: null               # Azure deployment name, e.g., "gpt-image-1.5" or "image-prod"
+  openai: null              # e.g., "gpt-image-2", "gpt-image-1.5", "gpt-image-1"
+  azure: null               # Azure deployment name, e.g., "gpt-image-2" or "image-prod"
  openrouter: null          # e.g., "google/gemini-3.1-flash-image-preview"
  dashscope: null           # e.g., "qwen-image-2.0-pro"
+  zai: null                 # e.g., "glm-image"
  minimax: null             # e.g., "image-01"
-  replicate: null           # e.g., "google/nano-banana-pro"
-  zai: null                 # e.g., "glm-image", "cogview-4-250304"
+  replicate: null           # e.g., "google/nano-banana-2"

 batch:
  max_workers: 10
@@ -50,10 +52,10 @@ batch:
    dashscope:
      concurrency: 3
      start_interval_ms: 1100
-    minimax:
+    zai:
      concurrency: 3
      start_interval_ms: 1100
-    zai:
+    minimax:
      concurrency: 3
      start_interval_ms: 1100
 ---
@@ -68,14 +70,15 @@ batch:
 | `default_quality` | string\|null | null | Default quality (null = 2k) |
 | `default_aspect_ratio` | string\|null | null | Default aspect ratio |
 | `default_image_size` | string\|null | null | Google/OpenRouter image size (overrides quality) |
+| `default_image_api_dialect` | string\|null | null | OpenAI-compatible image dialect (`openai-native` or `ratio-metadata`) |
 | `default_model.google` | string\|null | null | Google default model |
 | `default_model.openai` | string\|null | null | OpenAI default model |
 | `default_model.azure` | string\|null | null | Azure default deployment name |
 | `default_model.openrouter` | string\|null | null | OpenRouter default model |
 | `default_model.dashscope` | string\|null | null | DashScope default model |
+| `default_model.zai` | string\|null | null | Z.AI default model |
 | `default_model.minimax` | string\|null | null | MiniMax default model |
 | `default_model.replicate` | string\|null | null | Replicate default model |
-| `default_model.zai` | string\|null | null | Z.AI default model (glm-image / cogview-4-*) |
 | `batch.max_workers` | int\|null | 10 | Batch worker cap |
 | `batch.provider_limits.<provider>.concurrency` | int\|null | provider default | Max simultaneous requests per provider |
 | `batch.provider_limits.<provider>.start_interval_ms` | int\|null | provider default | Minimum gap between request starts per provider |
@@ -88,6 +91,7 @@ batch:
 version: 1
 default_provider: google
 default_quality: 2k
+default_image_api_dialect: null
 ---
 ```

@@ -99,15 +103,16 @@ default_provider: google
 default_quality: 2k
 default_aspect_ratio: "16:9"
 default_image_size: 2K
+default_image_api_dialect: null
 default_model:
  google: "gemini-3-pro-image-preview"
-  openai: "gpt-image-1.5"
-  azure: "gpt-image-1.5"
+  openai: "gpt-image-2"
+  azure: "gpt-image-2"
  openrouter: "google/gemini-3.1-flash-image-preview"
  dashscope: "qwen-image-2.0-pro"
-  minimax: "image-01"
-  replicate: "google/nano-banana-pro"
  zai: "glm-image"
+  minimax: "image-01"
+  replicate: "google/nano-banana-2"
 batch:
  max_workers: 10
  provider_limits:
@@ -117,14 +122,14 @@ batch:
    azure:
      concurrency: 3
      start_interval_ms: 1100
+    zai:
+      concurrency: 3
+      start_interval_ms: 1100
    openrouter:
      concurrency: 3
      start_interval_ms: 1100
    minimax:
      concurrency: 3
      start_interval_ms: 1100
-    zai:
-      concurrency: 3
-      start_interval_ms: 1100
 ---
 ```
@@ -17,6 +17,17 @@ Read when the user picks `--provider dashscope`, sets `default_model.dashscope`,
 - Default is `1664*928`
 - `qwen-image` currently has the same capability as `qwen-image-plus`

+**`wan2.7-image*`** — multimodal Wan 2.7 family. Members: `wan2.7-image-pro`, `wan2.7-image`.
+
+- Free-form `size` in `宽*高` format, plus aspect-ratio inference
+- `wan2.7-image-pro` text-to-image (no `--ref`): total pixels in `[768*768, 4096*4096]`, ratio in `[1:8, 8:1]`
+- `wan2.7-image-pro` with reference images and `wan2.7-image` (all scenarios): total pixels in `[768*768, 2048*2048]`, ratio in `[1:8, 8:1]`
+- Default: `1024*1024` (`--quality normal`) or `2048*2048` (`--quality 2k`); 4K requires explicit `--size`
+- Supports up to 9 reference images in `--ref` (image editing / multi-image fusion)
+- Reference images are sent inline as base64 (or passed through if the path is an `http(s)://` URL)
+- API does NOT use `prompt_extend`; the skill omits it for this family
+- The Wan 2.7 API defaults `n` to **4** in non-collage mode and bills per generated image. baoyu-image-gen forces `n: 1` and rejects `--n > 1` to avoid silently paying for and discarding extra images.
+
 **Legacy** — `z-image-turbo`, `z-image-ultra`, `wanx-v1`. Only use when the user explicitly asks for legacy behavior.

 ## Size Resolution
@@ -24,7 +35,8 @@ Read when the user picks `--provider dashscope`, sets `default_model.dashscope`,
 - `--size` wins over `--ar`
 - For `qwen-image-2.0*`: prefer explicit `--size`; otherwise infer from `--ar` using the recommended table below
 - For `qwen-image-max/plus/image`: only use the five fixed sizes; if the requested ratio doesn't fit, switch to `qwen-image-2.0-pro`
- `--quality` is a baoyu-imagine preset, not an official DashScope field. The mapping of `normal`/`2k` onto the `qwen-image-2.0*` table is an implementation choice, not an API guarantee
+- For `wan2.7-image*`: explicit `--size` is validated against the per-mode pixel/ratio limits; otherwise the size is derived from `--ar` and `--quality` (`normal` ≈ 1K, `2k` ≈ 2K). To request 4K with `wan2.7-image-pro` text-to-image, pass `--size` explicitly (e.g. `4096*4096`, `3840*2160`)
+- `--quality` is a baoyu-image-gen preset, not an official DashScope field. The mapping of `normal`/`2k` onto the `qwen-image-2.0*` and `wan2.7-image*` tables is an implementation choice, not an API guarantee

 ### Recommended `qwen-image-2.0*` sizes

@@ -39,12 +51,19 @@ Read when the user picks `--provider dashscope`, sets `default_model.dashscope`,
 | `16:9` | `1280*720` | `1920*1080` |
 | `21:9` | `1344*576` | `2048*872` |

+## Reference Images
+
+- Only `wan2.7-image-pro` and `wan2.7-image` accept `--ref`. Other DashScope models (qwen-image-2.0*, qwen-image-max/plus/image, legacy) reject `--ref` and the user is steered to a different provider/model.
+- Up to 9 reference images per request. Local files are inlined as base64 data URLs; `http(s)://` URLs are forwarded as-is.
+- Supplying any `--ref` automatically clamps the wan2.7-image-pro pixel ceiling from 4K to 2K (the API only supports 4K for pure text-to-image with no image input).
+
 ## Not Exposed

-DashScope APIs also support `negative_prompt`, `prompt_extend`, and `watermark`, but `baoyu-imagine` does not expose them as CLI flags today.
+DashScope APIs also support `negative_prompt`, `prompt_extend`, `watermark`, `thinking_mode`, `seed`, `bbox_list`, `enable_sequential`, and `color_palette`. `baoyu-image-gen` does not expose them as CLI flags today; the wan2.7 family relies on the API defaults (e.g. `thinking_mode=true`). The skill always sends `n=1` for wan2.7 — if you want grid/collage mode you currently need to call the API directly.

 ## Official References

 - [Qwen-Image API](https://help.aliyun.com/zh/model-studio/qwen-image-api)
 - [Text-to-image guide](https://help.aliyun.com/zh/model-studio/text-to-image)
 - [Qwen-Image Edit API](https://help.aliyun.com/zh/model-studio/qwen-image-edit-api)
+- [Wan 2.7 image generation & editing API](https://help.aliyun.com/zh/model-studio/wan-image-generation-and-editing-api-reference)
@@ -19,7 +19,7 @@ Read when the user picks `--provider minimax` or sets `default_model.minimax`. D

 - `--ref` files are sent as MiniMax `subject_reference`
 - `subject_reference[].type` is currently `character`
- Official docs say `image_file` supports public URLs or Base64 Data URLs; baoyu-imagine sends local refs as Data URLs
+- Official docs say `image_file` supports public URLs or Base64 Data URLs; baoyu-image-gen sends local refs as Data URLs
 - Recommended refs: front-facing portraits, JPG/JPEG/PNG, under 10MB

 ## Official References
@@ -1,6 +1,6 @@
 # Replicate

-Read when the user picks `--provider replicate`. Replicate support is intentionally scoped to model families baoyu-imagine can validate locally and save without dropping outputs.
+Read when the user picks `--provider replicate`. Replicate support is intentionally scoped to model families baoyu-image-gen can validate locally and save without dropping outputs.

 ## Supported Families

@@ -37,7 +37,7 @@ Read when the user picks `--provider replicate`. Replicate support is intentiona
 ## Guardrails

 - Replicate currently supports only single-output save semantics in this tool — keep `--n 1`
- If a model is outside the compatibility list above, baoyu-imagine treats it as prompt-only and rejects advanced local options instead of guessing a nano-banana-style schema
+- If a model is outside the compatibility list above, baoyu-image-gen treats it as prompt-only and rejects advanced local options instead of guessing a nano-banana-style schema

 ## Examples

@@ -6,7 +6,7 @@ Read when the user picks `--provider zai` or sets `default_model.zai`. Default m

 **`glm-image`** (recommended default)

- Text-to-image only in baoyu-imagine (no `--ref` support yet)
+- Text-to-image only in baoyu-image-gen (no `--ref` support yet)
 - Native `quality` options are `hd` and `standard`; this skill maps `2k → hd` and `normal → standard`
 - Recommended sizes: `1280x1280`, `1568x1056`, `1056x1568`, `1472x1088`, `1088x1472`, `1728x960`, `960x1728`
 - Custom `--size` requires width/height in `[1024, 2048]`, divisible by `32`, total pixels ≤ `2^22`
@@ -17,7 +17,7 @@ Read when the user picks `--provider zai` or sets `default_model.zai`. Default m

 ## Behavior Notes

- The sync API returns a temporary URL; baoyu-imagine downloads it and writes locally
+- The sync API returns a temporary URL; baoyu-image-gen downloads it and writes locally
 - `--ref` is not supported for Z.AI in this skill yet
 - The sync API returns a single image, so `--n > 1` is rejected

@@ -25,10 +25,13 @@ ${BUN_X} {baseDir}/scripts/main.ts --prompt "Make blue" --image out.png --ref so

 ```bash
 # OpenAI
-${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider openai
+${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider openai --model gpt-image-2

 # Azure OpenAI (model = deployment name)
-${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider azure --model gpt-image-1.5
+${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider azure --model gpt-image-2
+
+# OpenAI GPT Image 2 custom 4K size
+${BUN_X} {baseDir}/scripts/main.ts --prompt "A cinematic landscape" --image out.png --provider openai --model gpt-image-2 --size 3840x2160

 # Google with explicit model
 ${BUN_X} {baseDir}/scripts/main.ts --prompt "Make blue" --image out.png --provider google --model gemini-3-pro-image-preview --ref source.png
@@ -48,6 +51,12 @@ ${BUN_X} {baseDir}/scripts/main.ts --prompt "为咖啡品牌设计一张 21:9
 # DashScope legacy fixed-size
 ${BUN_X} {baseDir}/scripts/main.ts --prompt "一张电影感海报" --image out.png --provider dashscope --model qwen-image-max --size 1664x928

+# DashScope Wan 2.7 Image Pro (4K text-to-image)
+${BUN_X} {baseDir}/scripts/main.ts --prompt "一间有着精致窗户的花店" --image out.png --provider dashscope --model wan2.7-image-pro --size 4096x4096
+
+# DashScope Wan 2.7 Image with reference image (multi-image fusion)
+${BUN_X} {baseDir}/scripts/main.ts --prompt "把图2的涂鸦喷绘在图1的汽车上" --image out.png --provider dashscope --model wan2.7-image-pro --ref car.webp paint.webp
+
 # Z.AI GLM-image
 ${BUN_X} {baseDir}/scripts/main.ts --prompt "一张带清晰中文标题的科技海报" --image out.png --provider zai

@@ -0,0 +1,190 @@
+import assert from "node:assert/strict";
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+import { execFile } from "node:child_process";
+import { promisify } from "node:util";
+import test from "node:test";
+
+const execFileAsync = promisify(execFile);
+const repoRoot = path.resolve(import.meta.dirname, "..", "..", "..");
+const scriptPath = path.join(repoRoot, "skills", "baoyu-image-gen", "scripts", "build-batch.ts");
+
+async function makeFixture(): Promise<{
+  root: string;
+  outlinePath: string;
+  promptsDir: string;
+  outputPath: string;
+}> {
+  const root = await fs.mkdtemp(path.join(os.tmpdir(), "baoyu-image-gen-build-batch-"));
+  const outlinePath = path.join(root, "outline.md");
+  const promptsDir = path.join(root, "prompts");
+  const outputPath = path.join(root, "batch.json");
+
+  await fs.mkdir(promptsDir, { recursive: true });
+  await fs.writeFile(
+    outlinePath,
+    `## Illustration 1
+**Position**: demo
+**Purpose**: demo
+**Visual Content**: demo
+**Filename**: 01-demo.png
+`,
+  );
+  await fs.writeFile(path.join(promptsDir, "01-demo.md"), "A demo prompt\n");
+
+  return { root, outlinePath, promptsDir, outputPath };
+}
+
+async function runBuildBatch(args: string[]): Promise<void> {
+  await execFileAsync(process.execPath, ["--import", "tsx", scriptPath, ...args], {
+    cwd: repoRoot,
+  });
+}
+
+test("build-batch omits default model so baoyu-image-gen can resolve env or EXTEND defaults", async () => {
+  const fixture = await makeFixture();
+
+  await runBuildBatch([
+    "--outline",
+    fixture.outlinePath,
+    "--prompts",
+    fixture.promptsDir,
+    "--output",
+    fixture.outputPath,
+  ]);
+
+  const batch = JSON.parse(await fs.readFile(fixture.outputPath, "utf8")) as {
+    tasks: Array<Record<string, unknown>>;
+  };
+
+  assert.equal(batch.tasks.length, 1);
+  assert.equal(batch.tasks[0]?.provider, "replicate");
+  assert.equal(Object.hasOwn(batch.tasks[0]!, "model"), false);
+});
+
+test("build-batch preserves explicit model overrides", async () => {
+  const fixture = await makeFixture();
+
+  await runBuildBatch([
+    "--outline",
+    fixture.outlinePath,
+    "--prompts",
+    fixture.promptsDir,
+    "--output",
+    fixture.outputPath,
+    "--model",
+    "acme/custom-model",
+  ]);
+
+  const batch = JSON.parse(await fs.readFile(fixture.outputPath, "utf8")) as {
+    tasks: Array<Record<string, unknown>>;
+  };
+
+  assert.equal(batch.tasks[0]?.model, "acme/custom-model");
+});
+
+test("build-batch propagates direct-usage references from prompt frontmatter", async () => {
+  const fixture = await makeFixture();
+
+  await fs.writeFile(
+    path.join(fixture.promptsDir, "01-demo.md"),
+    `---
+illustration_id: 01
+type: infographic
+references:
+  - ref_id: 01
+    filename: 01-ref-brand.png
+    usage: direct
+  - ref_id: 02
+    filename: 02-ref-style.png
+    usage: style
+---
+
+A demo prompt
+`,
+  );
+
+  await runBuildBatch([
+    "--outline",
+    fixture.outlinePath,
+    "--prompts",
+    fixture.promptsDir,
+    "--output",
+    fixture.outputPath,
+  ]);
+
+  const batch = JSON.parse(await fs.readFile(fixture.outputPath, "utf8")) as {
+    tasks: Array<Record<string, unknown>>;
+  };
+
+  assert.deepEqual(batch.tasks[0]?.ref, ["references/01-ref-brand.png"]);
+});
+
+test("build-batch omits ref field when no direct references exist", async () => {
+  const fixture = await makeFixture();
+
+  await fs.writeFile(
+    path.join(fixture.promptsDir, "01-demo.md"),
+    `---
+illustration_id: 01
+references:
+  - ref_id: 01
+    filename: 01-ref-palette.png
+    usage: palette
+---
+
+A demo prompt
+`,
+  );
+
+  await runBuildBatch([
+    "--outline",
+    fixture.outlinePath,
+    "--prompts",
+    fixture.promptsDir,
+    "--output",
+    fixture.outputPath,
+  ]);
+
+  const batch = JSON.parse(await fs.readFile(fixture.outputPath, "utf8")) as {
+    tasks: Array<Record<string, unknown>>;
+  };
+
+  assert.equal(Object.hasOwn(batch.tasks[0]!, "ref"), false);
+});
+
+test("build-batch honors --refs-dir override", async () => {
+  const fixture = await makeFixture();
+
+  await fs.writeFile(
+    path.join(fixture.promptsDir, "01-demo.md"),
+    `---
+illustration_id: 01
+references:
+  - ref_id: 01
+    filename: brand.png
+    usage: direct
+---
+
+A demo prompt
+`,
+  );
+
+  await runBuildBatch([
+    "--outline",
+    fixture.outlinePath,
+    "--prompts",
+    fixture.promptsDir,
+    "--output",
+    fixture.outputPath,
+    "--refs-dir",
+    "refs",
+  ]);
+
+  const batch = JSON.parse(await fs.readFile(fixture.outputPath, "utf8")) as {
+    tasks: Array<Record<string, unknown>>;
+  };
+
+  assert.deepEqual(batch.tasks[0]?.ref, ["refs/brand.png"]);
+});
@@ -0,0 +1,238 @@
+import path from "node:path";
+import process from "node:process";
+import { readdir, readFile, writeFile } from "node:fs/promises";
+
+type CliArgs = {
+  outlinePath: string | null;
+  promptsDir: string | null;
+  outputPath: string | null;
+  imagesDir: string | null;
+  refsDir: string;
+  provider: string;
+  model: string | null;
+  aspectRatio: string;
+  quality: string;
+  jobs: number | null;
+  help: boolean;
+};
+
+type OutlineEntry = {
+  index: number;
+  filename: string;
+};
+
+type PromptReference = {
+  filename: string;
+  usage: "direct" | "style" | "palette";
+};
+
+function printUsage(): void {
+  console.log(`Usage:
+  npx -y tsx scripts/build-batch.ts --outline outline.md --prompts prompts --output batch.json --images-dir attachments
+
+Options:
+  --outline <path>     Path to outline.md
+  --prompts <path>     Path to prompts directory
+  --output <path>      Path to output batch.json
+  --images-dir <path>  Directory for generated images
+  --refs-dir <path>    Directory holding reference images, relative to batch file (default: references)
+  --provider <name>    Provider for baoyu-image-gen batch tasks (default: replicate)
+  --model <id>         Explicit model for baoyu-image-gen batch tasks (default: resolved by baoyu-image-gen config/env)
+  --ar <ratio>         Aspect ratio for all tasks (default: 16:9)
+  --quality <level>    Quality for all tasks (default: 2k)
+  --jobs <count>       Recommended worker count metadata (optional)
+  -h, --help           Show help`);
+}
+
+function parseArgs(argv: string[]): CliArgs {
+  const args: CliArgs = {
+    outlinePath: null,
+    promptsDir: null,
+    outputPath: null,
+    imagesDir: null,
+    refsDir: "references",
+    provider: "replicate",
+    model: null,
+    aspectRatio: "16:9",
+    quality: "2k",
+    jobs: null,
+    help: false,
+  };
+
+  for (let i = 0; i < argv.length; i++) {
+    const current = argv[i]!;
+    if (current === "--outline") args.outlinePath = argv[++i] ?? null;
+    else if (current === "--prompts") args.promptsDir = argv[++i] ?? null;
+    else if (current === "--output") args.outputPath = argv[++i] ?? null;
+    else if (current === "--images-dir") args.imagesDir = argv[++i] ?? null;
+    else if (current === "--refs-dir") args.refsDir = argv[++i] ?? args.refsDir;
+    else if (current === "--provider") args.provider = argv[++i] ?? args.provider;
+    else if (current === "--model") args.model = argv[++i] ?? args.model;
+    else if (current === "--ar") args.aspectRatio = argv[++i] ?? args.aspectRatio;
+    else if (current === "--quality") args.quality = argv[++i] ?? args.quality;
+    else if (current === "--jobs") {
+      const value = argv[++i];
+      args.jobs = value ? parseInt(value, 10) : null;
+    } else if (current === "--help" || current === "-h") {
+      args.help = true;
+    }
+  }
+  return args;
+}
+
+function parsePromptReferences(content: string): PromptReference[] {
+  const fmMatch = content.match(/^---\s*\n([\s\S]*?)\n---\s*(?:\n|$)/);
+  if (!fmMatch) return [];
+  const lines = fmMatch[1]!.split(/\r?\n/);
+
+  const refs: PromptReference[] = [];
+  let current: Partial<PromptReference> | null = null;
+  let inReferences = false;
+  let listIndent = 0;
+
+  const flush = () => {
+    if (current?.filename) {
+      refs.push({
+        filename: current.filename,
+        usage: (current.usage ?? "direct") as PromptReference["usage"],
+      });
+    }
+    current = null;
+  };
+
+  const unquote = (raw: string): string => raw.trim().replace(/^["']|["']$/g, "");
+
+  for (const line of lines) {
+    if (!line.trim() || line.trim().startsWith("#")) continue;
+
+    const keyMatch = line.match(/^(\S[^:]*):\s*(.*)$/);
+    if (keyMatch) {
+      flush();
+      if (keyMatch[1] === "references") {
+        inReferences = true;
+        listIndent = 0;
+        continue;
+      }
+      inReferences = false;
+      continue;
+    }
+
+    if (!inReferences) continue;
+
+    const itemMatch = line.match(/^(\s*)-\s*(.*)$/);
+    if (itemMatch) {
+      flush();
+      listIndent = itemMatch[1]!.length;
+      current = {};
+      const rest = itemMatch[2]!.trim();
+      if (rest) {
+        const kv = rest.match(/^(\w+)\s*:\s*(.*)$/);
+        if (kv && (kv[1] === "filename" || kv[1] === "usage")) {
+          (current as Record<string, string>)[kv[1]] = unquote(kv[2]!);
+        }
+      }
+      continue;
+    }
+
+    const kvMatch = line.match(/^(\s+)(\w+)\s*:\s*(.*)$/);
+    if (kvMatch && kvMatch[1]!.length > listIndent && current) {
+      if (kvMatch[2] === "filename" || kvMatch[2] === "usage") {
+        (current as Record<string, string>)[kvMatch[2]!] = unquote(kvMatch[3]!);
+      }
+    }
+  }
+  flush();
+
+  return refs;
+}
+
+function parseOutline(content: string): OutlineEntry[] {
+  const entries: OutlineEntry[] = [];
+  const blocks = content.split(/^## Illustration\s+/m).slice(1);
+
+  for (const block of blocks) {
+    const indexMatch = block.match(/^(\d+)/);
+    const filenameMatch = block.match(/\*\*Filename\*\*:\s*(.+)/);
+    if (indexMatch && filenameMatch) {
+      entries.push({
+        index: parseInt(indexMatch[1]!, 10),
+        filename: filenameMatch[1]!.trim(),
+      });
+    }
+  }
+  return entries;
+}
+
+async function findPromptFile(promptsDir: string, entry: OutlineEntry): Promise<string | null> {
+  const files = await readdir(promptsDir);
+  const prefix = String(entry.index).padStart(2, "0");
+  const match = files.find((f) => f.startsWith(prefix) && f.endsWith(".md"));
+  return match ? path.join(promptsDir, match) : null;
+}
+
+async function main(): Promise<void> {
+  const args = parseArgs(process.argv.slice(2));
+  if (args.help) {
+    printUsage();
+    return;
+  }
+
+  if (!args.outlinePath) {
+    console.error("Error: --outline is required");
+    process.exit(1);
+  }
+  if (!args.promptsDir) {
+    console.error("Error: --prompts is required");
+    process.exit(1);
+  }
+  if (!args.outputPath) {
+    console.error("Error: --output is required");
+    process.exit(1);
+  }
+
+  const outlineContent = await readFile(args.outlinePath, "utf8");
+  const entries = parseOutline(outlineContent);
+
+  if (entries.length === 0) {
+    console.error("No illustration entries found in outline.");
+    process.exit(1);
+  }
+
+  const tasks = [];
+  for (const entry of entries) {
+    const promptFile = await findPromptFile(args.promptsDir, entry);
+    if (!promptFile) {
+      console.error(`Warning: No prompt file found for illustration ${entry.index}, skipping.`);
+      continue;
+    }
+
+    const imageDir = args.imagesDir ?? path.dirname(args.outputPath);
+    const promptContent = await readFile(promptFile, "utf8");
+    const refs = parsePromptReferences(promptContent)
+      .filter((r) => r.usage === "direct")
+      .map((r) => path.posix.join(args.refsDir, r.filename));
+
+    const task: Record<string, unknown> = {
+      id: `illustration-${String(entry.index).padStart(2, "0")}`,
+      promptFiles: [promptFile],
+      image: path.join(imageDir, entry.filename),
+      provider: args.provider,
+      ar: args.aspectRatio,
+      quality: args.quality,
+    };
+    if (args.model) task.model = args.model;
+    if (refs.length > 0) task.ref = refs;
+    tasks.push(task);
+  }
+
+  const output: Record<string, unknown> = { tasks };
+  if (args.jobs) output.jobs = args.jobs;
+
+  await writeFile(args.outputPath, JSON.stringify(output, null, 2) + "\n");
+  console.log(`Batch file written: ${args.outputPath} (${tasks.length} tasks)`);
+}
+
+main().catch((error) => {
+  console.error(error instanceof Error ? error.message : String(error));
+  process.exit(1);
+});
@@ -13,10 +13,13 @@ import {
  getWorkerCount,
  isRetryableGenerationError,
  loadBatchTasks,
+  loadExtendConfig,
  mergeConfig,
  normalizeOutputImagePath,
  parseArgs,
+  parseOpenAIImageApiDialect,
  parseSimpleYaml,
+  validateReferenceImages,
 } from "./main.ts";

 function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
@@ -27,9 +30,12 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    provider: null,
    model: null,
    aspectRatio: null,
+    aspectRatioSource: null,
    size: null,
    quality: null,
    imageSize: null,
+    imageSizeSource: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -69,7 +75,7 @@ async function makeTempDir(prefix: string): Promise<string> {
  return fs.mkdtemp(path.join(os.tmpdir(), prefix));
 }

-test("parseArgs parses the main image-gen CLI flags", () => {
+test("parseArgs parses the main baoyu-image-gen CLI flags", () => {
  const args = parseArgs([
    "--promptfiles",
    "prompts/system.md",
@@ -77,11 +83,13 @@ test("parseArgs parses the main image-gen CLI flags", () => {
    "--image",
    "out/hero",
    "--provider",
-    "openai",
+    "zai",
    "--quality",
    "2k",
    "--imageSize",
    "4k",
+    "--imageApiDialect",
+    "ratio-metadata",
    "--ref",
    "ref/one.png",
    "ref/two.jpg",
@@ -94,9 +102,12 @@ test("parseArgs parses the main image-gen CLI flags", () => {

  assert.deepEqual(args.promptFiles, ["prompts/system.md", "prompts/content.md"]);
  assert.equal(args.imagePath, "out/hero");
-  assert.equal(args.provider, "openai");
+  assert.equal(args.provider, "zai");
  assert.equal(args.quality, "2k");
+  assert.equal(args.aspectRatioSource, null);
  assert.equal(args.imageSize, "4K");
+  assert.equal(args.imageSizeSource, "cli");
+  assert.equal(args.imageApiDialect, "ratio-metadata");
  assert.deepEqual(args.referenceImages, ["ref/one.png", "ref/two.jpg"]);
  assert.equal(args.n, 3);
  assert.equal(args.jobs, 5);
@@ -113,6 +124,15 @@ test("parseArgs falls back to positional prompt and rejects invalid provider", (
  );
 });

+test("validateReferenceImages can skip remote URLs for providers that support them", async () => {
+  await validateReferenceImages(["https://example.com/ref.png"], { allowRemoteUrls: true });
+
+  await assert.rejects(
+    () => validateReferenceImages(["https://example.com/ref.png"]),
+    /Reference image not found/,
+  );
+});
+
 test("parseSimpleYaml parses nested defaults and provider limits", () => {
  const yaml = `
 version: 2
@@ -120,9 +140,11 @@ default_provider: openrouter
 default_quality: normal
 default_aspect_ratio: '16:9'
 default_image_size: 2K
+default_image_api_dialect: ratio-metadata
 default_model:
  google: gemini-3-pro-image-preview
-  openai: gpt-image-1.5
+  openai: gpt-image-2
+  zai: glm-image
  azure: image-prod
  minimax: image-01
 batch:
@@ -133,6 +155,9 @@ batch:
      start_interval_ms: 900
    openai:
      concurrency: 4
+    zai:
+      concurrency: 2
+      start_interval_ms: 1000
    minimax:
      concurrency: 2
      start_interval_ms: 1400
@@ -148,8 +173,10 @@ batch:
  assert.equal(config.default_quality, "normal");
  assert.equal(config.default_aspect_ratio, "16:9");
  assert.equal(config.default_image_size, "2K");
+  assert.equal(config.default_image_api_dialect, "ratio-metadata");
  assert.equal(config.default_model?.google, "gemini-3-pro-image-preview");
-  assert.equal(config.default_model?.openai, "gpt-image-1.5");
+  assert.equal(config.default_model?.openai, "gpt-image-2");
+  assert.equal(config.default_model?.zai, "glm-image");
  assert.equal(config.default_model?.azure, "image-prod");
  assert.equal(config.default_model?.minimax, "image-01");
  assert.equal(config.batch?.max_workers, 8);
@@ -160,6 +187,10 @@ batch:
  assert.deepEqual(config.batch?.provider_limits?.openai, {
    concurrency: 4,
  });
+  assert.deepEqual(config.batch?.provider_limits?.zai, {
+    concurrency: 2,
+    start_interval_ms: 1000,
+  });
  assert.deepEqual(config.batch?.provider_limits?.minimax, {
    concurrency: 2,
    start_interval_ms: 1400,
@@ -170,6 +201,61 @@ batch:
  });
 });

+test("loadExtendConfig renames legacy EXTEND.md when the new path is missing", async () => {
+  const root = await makeTempDir("baoyu-image-gen-extend-");
+  const cwd = path.join(root, "project");
+  const home = path.join(root, "home");
+  const legacyPath = path.join(cwd, ".baoyu-skills", "baoyu-imagine", "EXTEND.md");
+  const currentPath = path.join(cwd, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md");
+
+  await fs.mkdir(path.dirname(legacyPath), { recursive: true });
+  await fs.mkdir(home, { recursive: true });
+  await fs.writeFile(legacyPath, `---
+default_provider: google
+default_quality: 2k
+---
+`);
+
+  const config = await loadExtendConfig(cwd, home);
+
+  assert.equal(config.default_provider, "google");
+  assert.equal(config.default_quality, "2k");
+  await fs.access(currentPath);
+  await assert.rejects(() => fs.access(legacyPath));
+});
+
+test("loadExtendConfig leaves legacy EXTEND.md untouched when both paths exist", async () => {
+  const root = await makeTempDir("baoyu-image-gen-extend-dual-");
+  const cwd = path.join(root, "project");
+  const home = path.join(root, "home");
+  const legacyPath = path.join(cwd, ".baoyu-skills", "baoyu-imagine", "EXTEND.md");
+  const currentPath = path.join(cwd, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md");
+
+  await fs.mkdir(path.dirname(legacyPath), { recursive: true });
+  await fs.mkdir(path.dirname(currentPath), { recursive: true });
+  await fs.mkdir(home, { recursive: true });
+  await fs.writeFile(legacyPath, `---
+default_provider: google
+---
+`);
+  await fs.writeFile(currentPath, `---
+default_provider: openai
+---
+`);
+
+  const config = await loadExtendConfig(cwd, home);
+
+  assert.equal(config.default_provider, "openai");
+  assert.equal(await fs.readFile(legacyPath, "utf8"), `---
+default_provider: google
+---
+`);
+  assert.equal(await fs.readFile(currentPath, "utf8"), `---
+default_provider: openai
+---
+`);
+});
+
 test("mergeConfig only fills values missing from CLI args", () => {
  const merged = mergeConfig(
    makeArgs({
@@ -183,13 +269,48 @@ test("mergeConfig only fills values missing from CLI args", () => {
      default_quality: "2k",
      default_aspect_ratio: "3:2",
      default_image_size: "2K",
+      default_image_api_dialect: "ratio-metadata",
    } satisfies Partial<ExtendConfig>,
  );

  assert.equal(merged.provider, "openai");
  assert.equal(merged.quality, "2k");
  assert.equal(merged.aspectRatio, "3:2");
+  assert.equal(merged.aspectRatioSource, "config");
  assert.equal(merged.imageSize, "4K");
+  assert.equal(merged.imageSizeSource, "cli");
+  assert.equal(merged.imageApiDialect, "ratio-metadata");
+});
+
+test("mergeConfig tags inherited imageSize defaults so providers can ignore incompatible config", () => {
+  const merged = mergeConfig(
+    makeArgs(),
+    {
+      default_image_size: "2K",
+    } satisfies Partial<ExtendConfig>,
+  );
+
+  assert.equal(merged.imageSize, "2K");
+  assert.equal(merged.imageSizeSource, "config");
+});
+
+test("mergeConfig falls back to OPENAI_IMAGE_API_DIALECT when CLI and EXTEND are unset", (t) => {
+  useEnv(t, {
+    OPENAI_IMAGE_API_DIALECT: "ratio-metadata",
+  });
+
+  const merged = mergeConfig(makeArgs(), {});
+  assert.equal(merged.imageApiDialect, "ratio-metadata");
+});
+
+test("parseOpenAIImageApiDialect validates supported values", () => {
+  assert.equal(parseOpenAIImageApiDialect("openai-native"), "openai-native");
+  assert.equal(parseOpenAIImageApiDialect("ratio-metadata"), "ratio-metadata");
+  assert.equal(parseOpenAIImageApiDialect(null), null);
+  assert.throws(
+    () => parseOpenAIImageApiDialect("gateway-magic"),
+    /Invalid OpenAI image API dialect/,
+  );
 });

 test("detectProvider rejects non-ref-capable providers and prefers Google first when multiple keys exist", (t) => {
@@ -197,7 +318,7 @@ test("detectProvider rejects non-ref-capable providers and prefers Google first
    () =>
      detectProvider(
        makeArgs({
-          provider: "dashscope",
+          provider: "zai",
          referenceImages: ["ref.png"],
        }),
      ),
@@ -260,6 +381,27 @@ test("detectProvider selects Azure when only Azure credentials are configured",
  );
 });

+test("detectProvider selects Z.AI when credentials are present or the model id matches", (t) => {
+  useEnv(t, {
+    GOOGLE_API_KEY: null,
+    OPENAI_API_KEY: null,
+    AZURE_OPENAI_API_KEY: null,
+    AZURE_OPENAI_BASE_URL: null,
+    OPENROUTER_API_KEY: null,
+    DASHSCOPE_API_KEY: null,
+    ZAI_API_KEY: "zai-key",
+    BIGMODEL_API_KEY: null,
+    MINIMAX_API_KEY: null,
+    REPLICATE_API_TOKEN: null,
+    JIMENG_ACCESS_KEY_ID: null,
+    JIMENG_SECRET_ACCESS_KEY: null,
+    ARK_API_KEY: null,
+  });
+
+  assert.equal(detectProvider(makeArgs()), "zai");
+  assert.equal(detectProvider(makeArgs({ model: "glm-image" })), "zai");
+});
+
 test("detectProvider infers Seedream from model id and allows Seedream reference-image workflows", (t) => {
  useEnv(t, {
    GOOGLE_API_KEY: null,
@@ -294,6 +436,33 @@ test("detectProvider infers Seedream from model id and allows Seedream reference
  );
 });

+test("detectProvider allows DashScope reference-image workflows when explicitly chosen for wan2.7 models", (t) => {
+  useEnv(t, {
+    GOOGLE_API_KEY: null,
+    OPENAI_API_KEY: null,
+    AZURE_OPENAI_API_KEY: null,
+    AZURE_OPENAI_BASE_URL: null,
+    OPENROUTER_API_KEY: null,
+    DASHSCOPE_API_KEY: "dashscope-key",
+    MINIMAX_API_KEY: null,
+    REPLICATE_API_TOKEN: null,
+    JIMENG_ACCESS_KEY_ID: null,
+    JIMENG_SECRET_ACCESS_KEY: null,
+    ARK_API_KEY: null,
+  });
+
+  assert.equal(
+    detectProvider(
+      makeArgs({
+        provider: "dashscope",
+        model: "wan2.7-image-pro",
+        referenceImages: ["ref.png"],
+      }),
+    ),
+    "dashscope",
+  );
+});
+
 test("detectProvider selects MiniMax when only MiniMax credentials are configured or the model id matches", (t) => {
  useEnv(t, {
    GOOGLE_API_KEY: null,
@@ -319,6 +488,7 @@ test("batch worker and provider-rate-limit configuration prefer env over EXTEND
    BAOYU_IMAGE_GEN_MAX_WORKERS: "12",
    BAOYU_IMAGE_GEN_GOOGLE_CONCURRENCY: "5",
    BAOYU_IMAGE_GEN_GOOGLE_START_INTERVAL_MS: "450",
+    BAOYU_IMAGE_GEN_ZAI_CONCURRENCY: "4",
  });

  const extendConfig: Partial<ExtendConfig> = {
@@ -329,6 +499,10 @@ test("batch worker and provider-rate-limit configuration prefer env over EXTEND
          concurrency: 2,
          start_interval_ms: 900,
        },
+        zai: {
+          concurrency: 1,
+          start_interval_ms: 1200,
+        },
        minimax: {
          concurrency: 1,
          start_interval_ms: 1500,
@@ -342,6 +516,10 @@ test("batch worker and provider-rate-limit configuration prefer env over EXTEND
    concurrency: 5,
    startIntervalMs: 450,
  });
+  assert.deepEqual(getConfiguredProviderRateLimits(extendConfig).zai, {
+    concurrency: 4,
+    startIntervalMs: 1200,
+  });
  assert.deepEqual(getConfiguredProviderRateLimits(extendConfig).minimax, {
    concurrency: 1,
    startIntervalMs: 1500,
@@ -363,7 +541,7 @@ test("loadBatchTasks and createTaskArgs resolve batch-relative paths", async (t)
          id: "hero",
          promptFiles: ["prompts/hero.md"],
          image: "out/hero",
-          ref: ["refs/hero.png"],
+          ref: ["refs/hero.png", "https://example.com/ref.png"],
          ar: "16:9",
        },
      ],
@@ -379,6 +557,7 @@ test("loadBatchTasks and createTaskArgs resolve batch-relative paths", async (t)
    makeArgs({
      provider: "replicate",
      quality: "2k",
+      imageApiDialect: "ratio-metadata",
      json: true,
    }),
    loaded.tasks[0]!,
@@ -391,10 +570,12 @@ test("loadBatchTasks and createTaskArgs resolve batch-relative paths", async (t)
  assert.equal(taskArgs.imagePath, path.join(loaded.batchDir, "out/hero"));
  assert.deepEqual(taskArgs.referenceImages, [
    path.join(loaded.batchDir, "refs/hero.png"),
+    "https://example.com/ref.png",
  ]);
  assert.equal(taskArgs.provider, "replicate");
  assert.equal(taskArgs.aspectRatio, "16:9");
  assert.equal(taskArgs.quality, "2k");
+  assert.equal(taskArgs.imageApiDialect, "ratio-metadata");
  assert.equal(taskArgs.json, true);
 });

@@ -408,5 +589,35 @@ test("path normalization, worker count, and retry classification follow expected
  assert.equal(getWorkerCount(5, 0, 4), 1);

  assert.equal(isRetryableGenerationError(new Error("API error (401): denied")), false);
+  assert.equal(
+    isRetryableGenerationError(
+      new Error("Replicate returned 2 outputs, but baoyu-image-gen currently supports saving exactly one image per request."),
+    ),
+    false,
+  );
+  assert.equal(
+    isRetryableGenerationError(
+      new Error("DashScope wan2.7 image models accept at most 9 reference images. Received 10."),
+    ),
+    false,
+  );
+  assert.equal(
+    isRetryableGenerationError(
+      new Error("DashScope wan2.7 image models in baoyu-image-gen support exactly one output image per request."),
+    ),
+    false,
+  );
+  assert.equal(
+    isRetryableGenerationError(
+      new Error("DashScope wan2.7 image models support aspect ratios in [1:8, 8:1]."),
+    ),
+    false,
+  );
+  assert.equal(
+    isRetryableGenerationError(
+      new Error("DashScope wan2.7-image requires total pixels between 768*768 and 2048*2048."),
+    ),
+    false,
+  );
  assert.equal(isRetryableGenerationError(new Error("socket hang up")), true);
 });
@@ -2,12 +2,13 @@ import path from "node:path";
 import process from "node:process";
 import { homedir } from "node:os";
 import { fileURLToPath } from "node:url";
-import { access, mkdir, readFile, writeFile } from "node:fs/promises";
+import { access, mkdir, readFile, rename, writeFile } from "node:fs/promises";
 import type {
  BatchFile,
  BatchTaskInput,
  CliArgs,
  ExtendConfig,
+  OpenAIImageApiDialect,
  Provider,
 } from "./types";

@@ -58,11 +59,11 @@ const DEFAULT_PROVIDER_RATE_LIMITS: Record<Provider, ProviderRateLimit> = {
  openai: { concurrency: 3, startIntervalMs: 1100 },
  openrouter: { concurrency: 3, startIntervalMs: 1100 },
  dashscope: { concurrency: 3, startIntervalMs: 1100 },
+  zai: { concurrency: 3, startIntervalMs: 1100 },
  minimax: { concurrency: 3, startIntervalMs: 1100 },
  jimeng: { concurrency: 3, startIntervalMs: 1100 },
  seedream: { concurrency: 3, startIntervalMs: 1100 },
  azure: { concurrency: 3, startIntervalMs: 1100 },
-  zai: { concurrency: 3, startIntervalMs: 1100 },
 };

 function printUsage(): void {
@@ -77,14 +78,15 @@ Options:
  --image <path>            Output image path (required in single-image mode)
  --batchfile <path>        JSON batch file for multi-image generation
  --jobs <count>            Worker count for batch mode (default: auto, max from config, built-in default 10)
-  --provider google|openai|openrouter|dashscope|minimax|replicate|jimeng|seedream|azure|zai  Force provider (auto-detect by default)
+  --provider google|openai|openrouter|dashscope|zai|minimax|replicate|jimeng|seedream|azure  Force provider (auto-detect by default)
  -m, --model <id>          Model ID
  --ar <ratio>              Aspect ratio (e.g., 16:9, 1:1, 4:3)
  --size <WxH>              Size (e.g., 1024x1024)
  --quality normal|2k       Quality preset (default: 2k)
  --imageSize 1K|2K|4K      Image size for Google/OpenRouter (default: from quality)
-  --ref <files...>          Reference images (Google, OpenAI, Azure, OpenRouter, Replicate, MiniMax, or Seedream 4.0/4.5/5.0)
-  --n <count>               Number of images for the current task (default: 1)
+  --imageApiDialect <id>    OpenAI-compatible image dialect: openai-native|ratio-metadata
+  --ref <files...>          Reference images (Google, OpenAI, Azure, OpenRouter, Replicate supported families, MiniMax, Seedream 4.0/4.5/5.0, or DashScope wan2.7-image*)
+  --n <count>               Number of images for the current task (default: 1; Replicate currently requires 1)
  --json                    JSON output
  -h, --help                Show help

@@ -97,7 +99,7 @@ Batch file format:
        "promptFiles": ["prompts/hero.md"],
        "image": "out/hero.png",
        "provider": "replicate",
-        "model": "google/nano-banana-pro",
+        "model": "google/nano-banana-2",
        "ar": "16:9"
      }
    ]
@@ -107,6 +109,7 @@ Behavior:
  - Batch mode automatically runs in parallel when pending tasks >= 2
  - Each image retries automatically up to 3 attempts
  - Batch summary reports success count, failure count, and per-image errors
+  - Replicate currently supports single-image save semantics only; --n must stay at 1

 Environment variables:
  OPENAI_API_KEY            OpenAI API key
@@ -114,30 +117,33 @@ Environment variables:
  GOOGLE_API_KEY            Google API key
  GEMINI_API_KEY            Gemini API key (alias for GOOGLE_API_KEY)
  DASHSCOPE_API_KEY         DashScope API key
+  ZAI_API_KEY               Z.AI API key
+  BIGMODEL_API_KEY          Backward-compatible alias for Z.AI API key
  MINIMAX_API_KEY           MiniMax API key
  REPLICATE_API_TOKEN       Replicate API token
  JIMENG_ACCESS_KEY_ID      Jimeng Access Key ID
  JIMENG_SECRET_ACCESS_KEY  Jimeng Secret Access Key
  ARK_API_KEY               Seedream/Ark API key
-  ZAI_API_KEY               Z.AI API key (alias: BIGMODEL_API_KEY)
-  BIGMODEL_API_KEY          Z.AI API key alias (legacy BigModel credentials)
-  OPENAI_IMAGE_MODEL        Default OpenAI model (gpt-image-1.5)
+  OPENAI_IMAGE_MODEL        Default OpenAI model (gpt-image-2)
  OPENROUTER_IMAGE_MODEL    Default OpenRouter model (google/gemini-3.1-flash-image-preview)
  GOOGLE_IMAGE_MODEL        Default Google model (gemini-3-pro-image-preview)
  DASHSCOPE_IMAGE_MODEL     Default DashScope model (qwen-image-2.0-pro)
+  ZAI_IMAGE_MODEL           Default Z.AI model (glm-image)
+  BIGMODEL_IMAGE_MODEL      Backward-compatible alias for Z.AI model (glm-image)
  MINIMAX_IMAGE_MODEL       Default MiniMax model (image-01)
-  REPLICATE_IMAGE_MODEL     Default Replicate model (google/nano-banana-pro)
+  REPLICATE_IMAGE_MODEL     Default Replicate model (google/nano-banana-2)
  JIMENG_IMAGE_MODEL        Default Jimeng model (jimeng_t2i_v40)
  SEEDREAM_IMAGE_MODEL      Default Seedream model (doubao-seedream-5-0-260128)
-  ZAI_IMAGE_MODEL           Default Z.AI model (glm-image)
-  BIGMODEL_IMAGE_MODEL      Z.AI model alias (legacy BigModel variable)
  OPENAI_BASE_URL           Custom OpenAI endpoint
+  OPENAI_IMAGE_API_DIALECT  OpenAI-compatible image dialect (openai-native|ratio-metadata)
  OPENAI_IMAGE_USE_CHAT     Use /chat/completions instead of /images/generations (true|false)
  OPENROUTER_BASE_URL       Custom OpenRouter endpoint
  OPENROUTER_HTTP_REFERER   Optional app URL for OpenRouter attribution
  OPENROUTER_TITLE          Optional app name for OpenRouter attribution
  GOOGLE_BASE_URL           Custom Google endpoint
  DASHSCOPE_BASE_URL        Custom DashScope endpoint
+  ZAI_BASE_URL              Custom Z.AI endpoint
+  BIGMODEL_BASE_URL         Backward-compatible alias for Z.AI endpoint
  MINIMAX_BASE_URL          Custom MiniMax endpoint
  REPLICATE_BASE_URL        Custom Replicate endpoint
  JIMENG_BASE_URL           Custom Jimeng endpoint
@@ -145,10 +151,8 @@ Environment variables:
  AZURE_OPENAI_BASE_URL     Azure OpenAI resource or deployment endpoint
  AZURE_OPENAI_DEPLOYMENT   Default Azure deployment name
  AZURE_API_VERSION         Azure API version (default: 2025-04-01-preview)
-  AZURE_OPENAI_IMAGE_MODEL  Backward-compatible Azure deployment/model alias (defaults to gpt-image-1.5)
+  AZURE_OPENAI_IMAGE_MODEL  Backward-compatible Azure deployment/model alias (defaults to gpt-image-2)
  SEEDREAM_BASE_URL         Custom Seedream endpoint
-  ZAI_BASE_URL              Custom Z.AI endpoint (defaults to https://api.z.ai/api/paas/v4)
-  BIGMODEL_BASE_URL         Z.AI endpoint alias (legacy BigModel variable)
  BAOYU_IMAGE_GEN_MAX_WORKERS  Override batch worker cap
  BAOYU_IMAGE_GEN_<PROVIDER>_CONCURRENCY  Override provider concurrency
  BAOYU_IMAGE_GEN_<PROVIDER>_START_INTERVAL_MS  Override provider start gap in ms
@@ -164,9 +168,12 @@ export function parseArgs(argv: string[]): CliArgs {
    provider: null,
    model: null,
    aspectRatio: null,
+    aspectRatioSource: null,
    size: null,
    quality: null,
    imageSize: null,
+    imageSizeSource: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -246,12 +253,12 @@ export function parseArgs(argv: string[]): CliArgs {
        v !== "openai" &&
        v !== "openrouter" &&
        v !== "dashscope" &&
+        v !== "zai" &&
        v !== "minimax" &&
        v !== "replicate" &&
        v !== "jimeng" &&
        v !== "seedream" &&
-        v !== "azure" &&
-        v !== "zai"
+        v !== "azure"
      ) {
        throw new Error(`Invalid provider: ${v}`);
      }
@@ -270,6 +277,7 @@ export function parseArgs(argv: string[]): CliArgs {
      const v = argv[++i];
      if (!v) throw new Error("Missing value for --ar");
      out.aspectRatio = v;
+      out.aspectRatioSource = "cli";
      continue;
    }

@@ -291,6 +299,16 @@ export function parseArgs(argv: string[]): CliArgs {
      const v = argv[++i]?.toUpperCase();
      if (v !== "1K" && v !== "2K" && v !== "4K") throw new Error(`Invalid imageSize: ${v}`);
      out.imageSize = v;
+      out.imageSizeSource = "cli";
+      continue;
+    }
+
+    if (a === "--imageApiDialect") {
+      const v = argv[++i];
+      if (v !== "openai-native" && v !== "ratio-metadata") {
+        throw new Error(`Invalid imageApiDialect: ${v}`);
+      }
+      out.imageApiDialect = v;
      continue;
    }

@@ -397,18 +415,21 @@ export function parseSimpleYaml(yaml: string): Partial<ExtendConfig> {
        config.default_aspect_ratio = cleaned === "null" ? null : cleaned;
      } else if (key === "default_image_size") {
        config.default_image_size = value === "null" ? null : value as "1K" | "2K" | "4K";
+      } else if (key === "default_image_api_dialect") {
+        config.default_image_api_dialect =
+          value === "null" ? null : parseOpenAIImageApiDialect(value);
      } else if (key === "default_model") {
        config.default_model = {
          google: null,
          openai: null,
          openrouter: null,
          dashscope: null,
+          zai: null,
          minimax: null,
          replicate: null,
          jimeng: null,
          seedream: null,
          azure: null,
-          zai: null,
        };
        currentKey = "default_model";
        currentProvider = null;
@@ -432,12 +453,12 @@ export function parseSimpleYaml(yaml: string): Partial<ExtendConfig> {
          key === "openai" ||
          key === "openrouter" ||
          key === "dashscope" ||
+          key === "zai" ||
          key === "minimax" ||
          key === "replicate" ||
          key === "jimeng" ||
          key === "seedream" ||
-          key === "azure" ||
-          key === "zai"
+          key === "azure"
        )
      ) {
        config.batch ??= {};
@@ -451,12 +472,12 @@ export function parseSimpleYaml(yaml: string): Partial<ExtendConfig> {
          key === "openai" ||
          key === "openrouter" ||
          key === "dashscope" ||
+          key === "zai" ||
          key === "minimax" ||
          key === "replicate" ||
          key === "jimeng" ||
          key === "seedream" ||
-          key === "azure" ||
-          key === "zai"
+          key === "azure"
        )
      ) {
        const cleaned = value.replace(/['"]/g, "");
@@ -482,14 +503,58 @@ export function parseSimpleYaml(yaml: string): Partial<ExtendConfig> {
  return config;
 }

-async function loadExtendConfig(): Promise<Partial<ExtendConfig>> {
-  const home = homedir();
-  const cwd = process.cwd();
+export function parseOpenAIImageApiDialect(
+  value: string | undefined | null
+): OpenAIImageApiDialect | null {
+  if (!value) return null;
+  const normalized = value.replace(/['"]/g, "").trim();
+  if (normalized === "openai-native" || normalized === "ratio-metadata") return normalized;
+  throw new Error(`Invalid OpenAI image API dialect: ${value}`);
+}

-  const paths = [
-    path.join(cwd, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md"),
-    path.join(home, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md"),
+type ExtendConfigPathPair = {
+  current: string;
+  legacy: string;
+};
+
+function getExtendConfigPathPairs(cwd: string, home: string): ExtendConfigPathPair[] {
+  return [
+    {
+      current: path.join(cwd, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md"),
+      legacy: path.join(cwd, ".baoyu-skills", "baoyu-imagine", "EXTEND.md"),
+    },
+    {
+      current: path.join(home, ".baoyu-skills", "baoyu-image-gen", "EXTEND.md"),
+      legacy: path.join(home, ".baoyu-skills", "baoyu-imagine", "EXTEND.md"),
+    },
  ];
+}
+
+async function exists(filePath: string): Promise<boolean> {
+  try {
+    await access(filePath);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+async function migrateLegacyExtendConfig(cwd: string, home: string): Promise<void> {
+  for (const { current, legacy } of getExtendConfigPathPairs(cwd, home)) {
+    const [hasCurrent, hasLegacy] = await Promise.all([exists(current), exists(legacy)]);
+    if (hasCurrent || !hasLegacy) continue;
+    await mkdir(path.dirname(current), { recursive: true });
+    await rename(legacy, current);
+  }
+}
+
+export async function loadExtendConfig(
+  cwd = process.cwd(),
+  home = homedir(),
+): Promise<Partial<ExtendConfig>> {
+  await migrateLegacyExtendConfig(cwd, home);
+
+  const paths = getExtendConfigPathPairs(cwd, home).map(({ current }) => current);

  for (const p of paths) {
    try {
@@ -506,12 +571,25 @@ async function loadExtendConfig(): Promise<Partial<ExtendConfig>> {
 }

 export function mergeConfig(args: CliArgs, extend: Partial<ExtendConfig>): CliArgs {
+  const aspectRatio = args.aspectRatio ?? extend.default_aspect_ratio ?? null;
+  const imageSize = args.imageSize ?? extend.default_image_size ?? null;
+  const imageApiDialect =
+    args.imageApiDialect ??
+    extend.default_image_api_dialect ??
+    parseOpenAIImageApiDialect(process.env.OPENAI_IMAGE_API_DIALECT);
  return {
    ...args,
    provider: args.provider ?? extend.default_provider ?? null,
    quality: args.quality ?? extend.default_quality ?? null,
-    aspectRatio: args.aspectRatio ?? extend.default_aspect_ratio ?? null,
-    imageSize: args.imageSize ?? extend.default_image_size ?? null,
+    aspectRatio,
+    aspectRatioSource:
+      args.aspectRatioSource ??
+      (args.aspectRatio !== null ? "cli" : (aspectRatio !== null ? "config" : null)),
+    imageSize,
+    imageSizeSource:
+      args.imageSizeSource ??
+      (args.imageSize !== null ? "cli" : (imageSize !== null ? "config" : null)),
+    imageApiDialect,
  };
 }

@@ -547,14 +625,14 @@ export function getConfiguredProviderRateLimits(
    openai: { ...DEFAULT_PROVIDER_RATE_LIMITS.openai },
    openrouter: { ...DEFAULT_PROVIDER_RATE_LIMITS.openrouter },
    dashscope: { ...DEFAULT_PROVIDER_RATE_LIMITS.dashscope },
+    zai: { ...DEFAULT_PROVIDER_RATE_LIMITS.zai },
    minimax: { ...DEFAULT_PROVIDER_RATE_LIMITS.minimax },
    jimeng: { ...DEFAULT_PROVIDER_RATE_LIMITS.jimeng },
    seedream: { ...DEFAULT_PROVIDER_RATE_LIMITS.seedream },
    azure: { ...DEFAULT_PROVIDER_RATE_LIMITS.azure },
-    zai: { ...DEFAULT_PROVIDER_RATE_LIMITS.zai },
  };

-  for (const provider of ["replicate", "google", "openai", "openrouter", "dashscope", "minimax", "jimeng", "seedream", "azure", "zai"] as Provider[]) {
+  for (const provider of ["replicate", "google", "openai", "openrouter", "dashscope", "zai", "minimax", "jimeng", "seedream", "azure"] as Provider[]) {
    const envPrefix = `BAOYU_IMAGE_GEN_${provider.toUpperCase()}`;
    const extendLimit = extendConfig.batch?.provider_limits?.[provider];
    configured[provider] = {
@@ -606,6 +684,7 @@ function inferProviderFromModel(model: string | null): Provider | null {
  const normalized = model.trim();
  if (normalized.includes("seedream") || normalized.includes("seededit")) return "seedream";
  if (normalized === "image-01" || normalized === "image-01-live") return "minimax";
+  if (normalized === "glm-image" || normalized === "cogview-4-250304") return "zai";
  return null;
 }

@@ -619,10 +698,11 @@ export function detectProvider(args: CliArgs): Provider {
    args.provider !== "openrouter" &&
    args.provider !== "replicate" &&
    args.provider !== "seedream" &&
-    args.provider !== "minimax"
+    args.provider !== "minimax" &&
+    args.provider !== "dashscope"
  ) {
    throw new Error(
-      "Reference images require a ref-capable provider. Use --provider google (Gemini multimodal), --provider openai (GPT Image edits), --provider azure (Azure OpenAI), --provider openrouter (OpenRouter multimodal), --provider replicate, --provider seedream for supported Seedream models, or --provider minimax for MiniMax subject-reference workflows."
+      "Reference images require a ref-capable provider. Use --provider google (Gemini multimodal), --provider openai (GPT Image edits), --provider azure (Azure OpenAI), --provider openrouter (OpenRouter multimodal), --provider replicate, --provider dashscope with a wan2.7 image model, --provider seedream for supported Seedream models, or --provider minimax for MiniMax subject-reference workflows."
    );
  }

@@ -633,11 +713,11 @@ export function detectProvider(args: CliArgs): Provider {
  const hasOpenai = !!process.env.OPENAI_API_KEY;
  const hasOpenrouter = !!process.env.OPENROUTER_API_KEY;
  const hasDashscope = !!process.env.DASHSCOPE_API_KEY;
+  const hasZai = !!(process.env.ZAI_API_KEY || process.env.BIGMODEL_API_KEY);
  const hasMinimax = !!process.env.MINIMAX_API_KEY;
  const hasReplicate = !!process.env.REPLICATE_API_TOKEN;
  const hasJimeng = !!(process.env.JIMENG_ACCESS_KEY_ID && process.env.JIMENG_SECRET_ACCESS_KEY);
  const hasSeedream = !!process.env.ARK_API_KEY;
-  const hasZai = !!(process.env.ZAI_API_KEY || process.env.BIGMODEL_API_KEY);
  const modelProvider = inferProviderFromModel(args.model);

  if (modelProvider === "seedream") {
@@ -654,6 +734,13 @@ export function detectProvider(args: CliArgs): Provider {
    return "minimax";
  }

+  if (modelProvider === "zai") {
+    if (!hasZai) {
+      throw new Error("Model looks like a Z.AI image model, but ZAI_API_KEY is not set.");
+    }
+    return "zai";
+  }
+
  if (args.referenceImages.length > 0) {
    if (hasGoogle) return "google";
    if (hasOpenai) return "openai";
@@ -673,24 +760,40 @@ export function detectProvider(args: CliArgs): Provider {
    hasAzure && "azure",
    hasOpenrouter && "openrouter",
    hasDashscope && "dashscope",
+    hasZai && "zai",
    hasMinimax && "minimax",
    hasReplicate && "replicate",
    hasJimeng && "jimeng",
    hasSeedream && "seedream",
-    hasZai && "zai",
  ].filter(Boolean) as Provider[];

  if (available.length === 1) return available[0]!;
  if (available.length > 1) return available[0]!;

  throw new Error(
-    "No API key found. Set GOOGLE_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, AZURE_OPENAI_API_KEY+AZURE_OPENAI_BASE_URL, OPENROUTER_API_KEY, DASHSCOPE_API_KEY, MINIMAX_API_KEY, REPLICATE_API_TOKEN, JIMENG keys, ARK_API_KEY, or ZAI_API_KEY/BIGMODEL_API_KEY.\n" +
+    "No API key found. Set GOOGLE_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, AZURE_OPENAI_API_KEY+AZURE_OPENAI_BASE_URL, OPENROUTER_API_KEY, DASHSCOPE_API_KEY, ZAI_API_KEY, MINIMAX_API_KEY, REPLICATE_API_TOKEN, JIMENG keys, or ARK_API_KEY.\n" +
      "Create ~/.baoyu-skills/.env or <cwd>/.baoyu-skills/.env with your keys."
  );
 }

-export async function validateReferenceImages(referenceImages: string[]): Promise<void> {
+export type ReferenceImageValidationOptions = {
+  allowRemoteUrls?: boolean;
+};
+
+function isRemoteReferenceImage(refPath: string): boolean {
+  return /^https?:\/\//i.test(refPath);
+}
+
+function shouldAllowRemoteReferenceImages(provider: Provider | null): boolean {
+  return provider === "dashscope";
+}
+
+export async function validateReferenceImages(
+  referenceImages: string[],
+  options: ReferenceImageValidationOptions = {},
+): Promise<void> {
  for (const refPath of referenceImages) {
+    if (options.allowRemoteUrls && isRemoteReferenceImage(refPath)) continue;
    const fullPath = path.resolve(refPath);
    try {
      await access(fullPath);
@@ -716,6 +819,12 @@ export function isRetryableGenerationError(error: unknown): boolean {
    "API error (403)",
    "API error (404)",
    "temporarily disabled",
+    "supports saving exactly one image",
+    "supports only",
+    "support exactly one output image",
+    "support aspect ratios in",
+    "requires total pixels between",
+    "accept at most",
  ];
  return !nonRetryableMarkers.some((marker) => msg.includes(marker));
 }
@@ -723,13 +832,13 @@ export function isRetryableGenerationError(error: unknown): boolean {
 async function loadProviderModule(provider: Provider): Promise<ProviderModule> {
  if (provider === "google") return (await import("./providers/google")) as ProviderModule;
  if (provider === "dashscope") return (await import("./providers/dashscope")) as ProviderModule;
+  if (provider === "zai") return (await import("./providers/zai")) as ProviderModule;
  if (provider === "minimax") return (await import("./providers/minimax")) as ProviderModule;
  if (provider === "replicate") return (await import("./providers/replicate")) as ProviderModule;
  if (provider === "openrouter") return (await import("./providers/openrouter")) as ProviderModule;
  if (provider === "jimeng") return (await import("./providers/jimeng")) as ProviderModule;
  if (provider === "seedream") return (await import("./providers/seedream")) as ProviderModule;
  if (provider === "azure") return (await import("./providers/azure")) as ProviderModule;
-  if (provider === "zai") return (await import("./providers/zai")) as ProviderModule;
  return (await import("./providers/openai")) as ProviderModule;
 }

@@ -755,12 +864,12 @@ function getModelForProvider(
      return extendConfig.default_model.openrouter;
    }
    if (provider === "dashscope" && extendConfig.default_model.dashscope) return extendConfig.default_model.dashscope;
+    if (provider === "zai" && extendConfig.default_model.zai) return extendConfig.default_model.zai;
    if (provider === "minimax" && extendConfig.default_model.minimax) return extendConfig.default_model.minimax;
    if (provider === "replicate" && extendConfig.default_model.replicate) return extendConfig.default_model.replicate;
    if (provider === "jimeng" && extendConfig.default_model.jimeng) return extendConfig.default_model.jimeng;
    if (provider === "seedream" && extendConfig.default_model.seedream) return extendConfig.default_model.seedream;
    if (provider === "azure" && extendConfig.default_model.azure) return extendConfig.default_model.azure;
-    if (provider === "zai" && extendConfig.default_model.zai) return extendConfig.default_model.zai;
  }
  return providerModule.getDefaultModel();
 }
@@ -771,7 +880,11 @@ async function prepareSingleTask(args: CliArgs, extendConfig: Partial<ExtendConf
  const prompt = (await loadPromptForArgs(args)) ?? (await readPromptFromStdin());
  if (!prompt) throw new Error("Prompt is required");
  if (!args.imagePath) throw new Error("--image is required");
-  if (args.referenceImages.length > 0) await validateReferenceImages(args.referenceImages);
+  if (args.referenceImages.length > 0) {
+    await validateReferenceImages(args.referenceImages, {
+      allowRemoteUrls: shouldAllowRemoteReferenceImages(args.provider),
+    });
+  }

  const provider = detectProvider(args);
  const providerModule = await loadProviderModule(provider);
@@ -820,6 +933,10 @@ export function resolveBatchPath(batchDir: string, filePath: string): string {
  return path.isAbsolute(filePath) ? filePath : path.resolve(batchDir, filePath);
 }

+function resolveBatchReferencePath(batchDir: string, filePath: string): string {
+  return isRemoteReferenceImage(filePath) ? filePath : resolveBatchPath(batchDir, filePath);
+}
+
 export function createTaskArgs(baseArgs: CliArgs, task: BatchTaskInput, batchDir: string): CliArgs {
  return {
    ...baseArgs,
@@ -829,10 +946,13 @@ export function createTaskArgs(baseArgs: CliArgs, task: BatchTaskInput, batchDir
    provider: task.provider ?? baseArgs.provider ?? null,
    model: task.model ?? baseArgs.model ?? null,
    aspectRatio: task.ar ?? baseArgs.aspectRatio ?? null,
+    aspectRatioSource: task.ar != null ? "task" : (baseArgs.aspectRatioSource ?? null),
    size: task.size ?? baseArgs.size ?? null,
    quality: task.quality ?? baseArgs.quality ?? null,
    imageSize: task.imageSize ?? baseArgs.imageSize ?? null,
-    referenceImages: task.ref ? task.ref.map((filePath) => resolveBatchPath(batchDir, filePath)) : [],
+    imageSizeSource: task.imageSize != null ? "task" : (baseArgs.imageSizeSource ?? null),
+    imageApiDialect: task.imageApiDialect ?? baseArgs.imageApiDialect ?? null,
+    referenceImages: task.ref ? task.ref.map((filePath) => resolveBatchReferencePath(batchDir, filePath)) : [],
    n: task.n ?? baseArgs.n,
    batchFile: null,
    jobs: baseArgs.jobs,
@@ -856,7 +976,11 @@ async function prepareBatchTasks(
    const prompt = await loadPromptForArgs(taskArgs);
    if (!prompt) throw new Error(`Task ${i + 1} is missing prompt or promptFiles.`);
    if (!taskArgs.imagePath) throw new Error(`Task ${i + 1} is missing image output path.`);
-    if (taskArgs.referenceImages.length > 0) await validateReferenceImages(taskArgs.referenceImages);
+    if (taskArgs.referenceImages.length > 0) {
+      await validateReferenceImages(taskArgs.referenceImages, {
+        allowRemoteUrls: shouldAllowRemoteReferenceImages(taskArgs.provider),
+      });
+    }

    const provider = detectProvider(taskArgs);
    const providerModule = await loadProviderModule(provider);
@@ -980,7 +1104,7 @@ async function runBatchTasks(
  const acquireProvider = createProviderGate(providerRateLimits);
  const workerCount = getWorkerCount(tasks.length, jobs, maxWorkers);
  console.error(`Batch mode: ${tasks.length} tasks, ${workerCount} workers, parallel mode enabled.`);
-  for (const provider of ["replicate", "google", "openai", "openrouter", "dashscope", "jimeng", "seedream", "azure", "zai"] as Provider[]) {
+  for (const provider of ["replicate", "google", "openai", "openrouter", "dashscope", "zai", "minimax", "jimeng", "seedream", "azure"] as Provider[]) {
    const limit = providerRateLimits[provider];
    console.error(`- ${provider}: concurrency=${limit.concurrency}, startIntervalMs=${limit.startIntervalMs}`);
  }
@@ -48,6 +48,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -46,7 +46,7 @@ export function getDefaultModel(): string {
    }
  }

-  return process.env.AZURE_OPENAI_IMAGE_MODEL || "gpt-image-1.5";
+  return process.env.AZURE_OPENAI_IMAGE_MODEL || "gpt-image-2";
 }

 function getEndpoint(): AzureEndpoint {
@@ -2,15 +2,42 @@ import assert from "node:assert/strict";
 import test, { type TestContext } from "node:test";

 import {
+  generateImage,
  getDefaultModel,
  getModelFamily,
  getQwen2SizeFromAspectRatio,
  getSizeFromAspectRatio,
+  getWan27SizeFromAspectRatio,
  normalizeSize,
  parseAspectRatio,
  parseSize,
  resolveSizeForModel,
 } from "./dashscope.ts";
+import type { CliArgs } from "../types.ts";
+
+function makeCliArgs(overrides: Partial<CliArgs> = {}): CliArgs {
+  return {
+    prompt: null,
+    promptFiles: [],
+    imagePath: null,
+    provider: "dashscope",
+    model: null,
+    aspectRatio: null,
+    aspectRatioSource: null,
+    size: null,
+    quality: "2k",
+    imageSize: null,
+    imageSizeSource: null,
+    imageApiDialect: null,
+    referenceImages: [],
+    n: 1,
+    batchFile: null,
+    jobs: null,
+    json: false,
+    help: false,
+    ...overrides,
+  };
+}

 function useEnv(
  t: TestContext,
@@ -51,9 +78,11 @@ test("DashScope aspect-ratio parsing accepts numeric ratios only", () => {
  assert.equal(parseAspectRatio("-1:2"), null);
 });

-test("DashScope model family routing distinguishes qwen-2.0, fixed-size qwen, and legacy models", () => {
+test("DashScope model family routing distinguishes qwen-2.0, fixed-size qwen, wan2.7, and legacy models", () => {
  assert.equal(getModelFamily("qwen-image-2.0-pro"), "qwen2");
  assert.equal(getModelFamily("qwen-image"), "qwenFixed");
+  assert.equal(getModelFamily("wan2.7-image"), "wan27");
+  assert.equal(getModelFamily("wan2.7-image-pro"), "wan27");
  assert.equal(getModelFamily("z-image-turbo"), "legacy");
  assert.equal(getModelFamily("wanx-v1"), "legacy");
 });
@@ -146,3 +175,218 @@ test("DashScope size normalization converts WxH into provider format", () => {
  assert.equal(normalizeSize("1024x1024"), "1024*1024");
  assert.equal(normalizeSize("2048*1152"), "2048*1152");
 });
+
+test("Wan 2.7 derives sizes that match the requested ratio at the chosen pixel budget", () => {
+  const square2k = getWan27SizeFromAspectRatio(null, "2k", 2048 * 2048);
+  const parsedSquare = parseSize(square2k);
+  assert.ok(parsedSquare);
+  assert.equal(parsedSquare.width, parsedSquare.height);
+  assert.ok(parsedSquare.width * parsedSquare.height <= 2048 * 2048);
+
+  const widescreen = getWan27SizeFromAspectRatio("16:9", "2k", 2048 * 2048);
+  const parsedWide = parseSize(widescreen);
+  assert.ok(parsedWide);
+  assert.ok(Math.abs(parsedWide.width / parsedWide.height - 16 / 9) < 0.05);
+  assert.ok(parsedWide.width * parsedWide.height <= 2048 * 2048);
+
+  const pro4k = getWan27SizeFromAspectRatio("16:9", "2k", 4096 * 4096);
+  const parsed4k = parseSize(pro4k);
+  assert.ok(parsed4k);
+  assert.ok(parsed4k.width * parsed4k.height > 2048 * 2048);
+  assert.ok(parsed4k.width * parsed4k.height <= 4096 * 4096);
+});
+
+test("Wan 2.7 rejects aspect ratios outside the [1:8, 8:1] range", () => {
+  assert.throws(
+    () => getWan27SizeFromAspectRatio("9:1", "2k", 2048 * 2048),
+    /1:8, 8:1/,
+  );
+  assert.throws(
+    () => getWan27SizeFromAspectRatio("1:9", "normal", 2048 * 2048),
+    /1:8, 8:1/,
+  );
+});
+
+test("Wan 2.7 derived sizes stay inside the boundary ratio limits after rounding", () => {
+  for (const ar of ["8:1", "1:8"]) {
+    const size = getWan27SizeFromAspectRatio(ar, "2k", 2048 * 2048);
+    const parsed = parseSize(size);
+    assert.ok(parsed);
+    const ratio = parsed.width / parsed.height;
+    assert.ok(ratio >= 1 / 8);
+    assert.ok(ratio <= 8);
+    assert.ok(parsed.width * parsed.height <= 2048 * 2048);
+  }
+});
+
+test("resolveSizeForModel routes wan2.7-image to the 2K-capped derivation", () => {
+  const size = resolveSizeForModel("wan2.7-image", {
+    size: null,
+    aspectRatio: "16:9",
+    quality: "2k",
+  });
+  const parsed = parseSize(size);
+  assert.ok(parsed);
+  assert.ok(parsed.width * parsed.height <= 2048 * 2048);
+  assert.ok(Math.abs(parsed.width / parsed.height - 16 / 9) < 0.05);
+});
+
+test("resolveSizeForModel allows wan2.7-image-pro 4K only when there are no reference images", () => {
+  assert.equal(
+    resolveSizeForModel("wan2.7-image-pro", {
+      size: "4096*4096",
+      aspectRatio: null,
+      quality: "2k",
+    }),
+    "4096*4096",
+  );
+
+  assert.throws(
+    () =>
+      resolveSizeForModel("wan2.7-image-pro", {
+        size: "4096*4096",
+        aspectRatio: null,
+        quality: "2k",
+        referenceImages: ["a.png"],
+      }),
+    /total pixels between 768\*768 and 2048\*2048/,
+  );
+
+  const proWithRef = resolveSizeForModel("wan2.7-image-pro", {
+    size: null,
+    aspectRatio: "1:1",
+    quality: "2k",
+    referenceImages: ["a.png"],
+  });
+  const parsedRef = parseSize(proWithRef);
+  assert.ok(parsedRef);
+  assert.ok(parsedRef.width * parsedRef.height <= 2048 * 2048);
+});
+
+test("Wan 2.7 request body forces n=1 and omits prompt_extend / negative_prompt", async (t) => {
+  useEnv(t, { DASHSCOPE_API_KEY: "fake-key" });
+
+  const originalFetch = globalThis.fetch;
+  let capturedBody: any = null;
+  globalThis.fetch = (async (_url: string, init?: RequestInit) => {
+    capturedBody = JSON.parse(String(init?.body));
+    return new Response(
+      JSON.stringify({
+        output: {
+          choices: [
+            {
+              message: {
+                content: [{ image: "data:image/png;base64,iVBORw0KGgo=" }],
+              },
+            },
+          ],
+        },
+      }),
+      { status: 200, headers: { "content-type": "application/json" } },
+    );
+  }) as typeof fetch;
+  t.after(() => {
+    globalThis.fetch = originalFetch;
+  });
+
+  await generateImage("hello", "wan2.7-image-pro", makeCliArgs({ aspectRatio: "1:1" }));
+
+  assert.equal(capturedBody.model, "wan2.7-image-pro");
+  assert.deepEqual(Object.keys(capturedBody.parameters).sort(), ["n", "size", "watermark"]);
+  assert.equal(capturedBody.parameters.n, 1);
+  assert.equal(capturedBody.parameters.watermark, false);
+  assert.equal(typeof capturedBody.parameters.size, "string");
+  assert.ok(!("prompt_extend" in capturedBody.parameters));
+  assert.ok(!("negative_prompt" in capturedBody.parameters));
+
+  assert.deepEqual(capturedBody.input.messages[0].content, [{ text: "hello" }]);
+});
+
+test("Wan 2.7 request body forwards remote reference image URLs", async (t) => {
+  useEnv(t, { DASHSCOPE_API_KEY: "fake-key" });
+
+  const originalFetch = globalThis.fetch;
+  let capturedBody: any = null;
+  globalThis.fetch = (async (_url: string, init?: RequestInit) => {
+    capturedBody = JSON.parse(String(init?.body));
+    return new Response(
+      JSON.stringify({
+        output: {
+          choices: [
+            {
+              message: {
+                content: [{ image: "data:image/png;base64,iVBORw0KGgo=" }],
+              },
+            },
+          ],
+        },
+      }),
+      { status: 200, headers: { "content-type": "application/json" } },
+    );
+  }) as typeof fetch;
+  t.after(() => {
+    globalThis.fetch = originalFetch;
+  });
+
+  await generateImage(
+    "combine these",
+    "wan2.7-image-pro",
+    makeCliArgs({ referenceImages: ["https://example.com/ref.png"] }),
+  );
+
+  assert.deepEqual(capturedBody.input.messages[0].content, [
+    { image: "https://example.com/ref.png" },
+    { text: "combine these" },
+  ]);
+});
+
+test("Wan 2.7 rejects --n > 1 to prevent silent multi-image billing", async (t) => {
+  useEnv(t, { DASHSCOPE_API_KEY: "fake-key" });
+
+  await assert.rejects(
+    () => generateImage("hi", "wan2.7-image-pro", makeCliArgs({ n: 2 })),
+    /support exactly one output image/,
+  );
+});
+
+test("resolveSizeForModel validates explicit wan2.7 sizes by pixel budget and ratio", () => {
+  assert.equal(
+    resolveSizeForModel("wan2.7-image-pro", {
+      size: "3840x2160",
+      aspectRatio: null,
+      quality: "2k",
+    }),
+    "3840*2160",
+  );
+
+  assert.throws(
+    () =>
+      resolveSizeForModel("wan2.7-image-pro", {
+        size: "3840x2160",
+        aspectRatio: null,
+        quality: "2k",
+        referenceImages: ["a.png"],
+      }),
+    /total pixels between 768\*768 and 2048\*2048/,
+  );
+
+  assert.throws(
+    () =>
+      resolveSizeForModel("wan2.7-image", {
+        size: "4096x4096",
+        aspectRatio: null,
+        quality: "2k",
+      }),
+    /total pixels between 768\*768 and 2048\*2048/,
+  );
+
+  assert.throws(
+    () =>
+      resolveSizeForModel("wan2.7-image-pro", {
+        size: "3072*256",
+        aspectRatio: null,
+        quality: "2k",
+      }),
+    /1:8, 8:1/,
+  );
+});
@@ -1,6 +1,8 @@
+import path from "node:path";
+import { readFile } from "node:fs/promises";
 import type { CliArgs, Quality } from "../types";

-type DashScopeModelFamily = "qwen2" | "qwenFixed" | "legacy";
+type DashScopeModelFamily = "qwen2" | "qwenFixed" | "wan27" | "legacy";

 type DashScopeModelSpec = {
  family: DashScopeModelFamily;
@@ -19,6 +21,16 @@ const QWEN_2_TARGET_PIXELS: Record<Quality, number> = {
  "2k": 1536 * 1536,
 };

+const MIN_WAN27_TOTAL_PIXELS = 768 * 768;
+const MAX_WAN27_PRO_T2I_PIXELS = 4096 * 4096;
+const MAX_WAN27_GENERAL_PIXELS = 2048 * 2048;
+const WAN27_MAX_REFERENCE_IMAGES = 9;
+
+const WAN27_TARGET_PIXELS: Record<Quality, number> = {
+  normal: 1024 * 1024,
+  "2k": 2048 * 2048,
+};
+
 const QWEN_2_RECOMMENDED: Record<string, Record<Quality, string>> = {
  "1:1": { normal: "1024*1024", "2k": "1536*1536" },
  "2:3": { normal: "768*1152", "2k": "1024*1536" },
@@ -73,6 +85,11 @@ const QWEN_FIXED_SPEC: DashScopeModelSpec = {
  defaultSize: QWEN_FIXED_SIZES_BY_RATIO["16:9"],
 };

+const WAN27_SPEC: DashScopeModelSpec = {
+  family: "wan27",
+  defaultSize: "2048*2048",
+};
+
 const LEGACY_SPEC: DashScopeModelSpec = {
  family: "legacy",
  defaultSize: "1536*1536",
@@ -88,12 +105,31 @@ const MODEL_SPEC_ALIASES: Record<string, DashScopeModelSpec> = {
  "qwen-image-plus": QWEN_FIXED_SPEC,
  "qwen-image-plus-2026-01-09": QWEN_FIXED_SPEC,
  "qwen-image": QWEN_FIXED_SPEC,
+  "wan2.7-image-pro": WAN27_SPEC,
+  "wan2.7-image": WAN27_SPEC,
 };

 export function getDefaultModel(): string {
  return process.env.DASHSCOPE_IMAGE_MODEL || DEFAULT_MODEL;
 }

+function getReferenceImageMime(filePath: string): string {
+  const ext = path.extname(filePath).toLowerCase();
+  if (ext === ".jpg" || ext === ".jpeg") return "image/jpeg";
+  if (ext === ".webp") return "image/webp";
+  if (ext === ".bmp") return "image/bmp";
+  return "image/png";
+}
+
+async function loadReferenceImage(refPath: string): Promise<string> {
+  if (/^https?:\/\//i.test(refPath)) {
+    return refPath;
+  }
+  const fullPath = path.resolve(refPath);
+  const bytes = await readFile(fullPath);
+  return `data:${getReferenceImageMime(fullPath)};base64,${bytes.toString("base64")}`;
+}
+
 function getApiKey(): string | null {
  return process.env.DASHSCOPE_API_KEY || null;
 }
@@ -173,6 +209,10 @@ function roundToStep(value: number): number {
  return Math.max(SIZE_STEP, Math.round(value / SIZE_STEP) * SIZE_STEP);
 }

+function floorToStep(value: number): number {
+  return Math.max(SIZE_STEP, Math.floor(value / SIZE_STEP) * SIZE_STEP);
+}
+
 function fitToPixelBudget(
  width: number,
  height: number,
@@ -220,6 +260,21 @@ function fitToPixelBudget(
  return { width: roundedWidth, height: roundedHeight };
 }

+function clampWan27DerivedSizeToRatioBounds(
+  size: { width: number; height: number },
+): { width: number; height: number } {
+  let { width, height } = size;
+  const ratio = width / height;
+
+  if (ratio > 8) {
+    width = floorToStep(height * 8);
+  } else if (ratio < 1 / 8) {
+    height = floorToStep(width * 8);
+  }
+
+  return { width, height };
+}
+
 export function getSizeFromAspectRatio(ar: string | null, quality: CliArgs["quality"]): string {
  const normalizedQuality = normalizeQuality(quality);
  const sizes = normalizedQuality === "2k" ? LEGACY_STANDARD_SIZES_2K : LEGACY_STANDARD_SIZES;
@@ -276,6 +331,77 @@ export function getQwen2SizeFromAspectRatio(ar: string | null, quality: CliArgs[
  return formatSize(fitted.width, fitted.height);
 }

+function isWan27ProModel(model: string): boolean {
+  return model.trim().toLowerCase() === "wan2.7-image-pro";
+}
+
+function getWan27MaxPixels(model: string, hasReferenceImages: boolean): number {
+  if (isWan27ProModel(model) && !hasReferenceImages) {
+    return MAX_WAN27_PRO_T2I_PIXELS;
+  }
+  return MAX_WAN27_GENERAL_PIXELS;
+}
+
+export function getWan27SizeFromAspectRatio(
+  ar: string | null,
+  quality: CliArgs["quality"],
+  maxPixels: number,
+): string {
+  const normalizedQuality = normalizeQuality(quality);
+  const targetPixels = Math.min(WAN27_TARGET_PIXELS[normalizedQuality], maxPixels);
+
+  if (!ar) {
+    const side = roundToStep(Math.sqrt(targetPixels));
+    return formatSize(side, side);
+  }
+
+  const parsed = parseAspectRatio(ar);
+  if (!parsed) {
+    const side = roundToStep(Math.sqrt(targetPixels));
+    return formatSize(side, side);
+  }
+
+  const ratio = parsed.width / parsed.height;
+  if (ratio < 1 / 8 || ratio > 8) {
+    throw new Error(
+      `DashScope wan2.7 image models support aspect ratios in [1:8, 8:1]. Received "${ar}".`
+    );
+  }
+
+  const rawWidth = Math.sqrt(targetPixels * ratio);
+  const rawHeight = Math.sqrt(targetPixels / ratio);
+  const fitted = fitToPixelBudget(
+    rawWidth,
+    rawHeight,
+    MIN_WAN27_TOTAL_PIXELS,
+    maxPixels,
+  );
+  const bounded = clampWan27DerivedSizeToRatioBounds(fitted);
+
+  return formatSize(bounded.width, bounded.height);
+}
+
+function validateWan27Size(size: string, maxPixels: number, model: string): string {
+  const normalized = normalizeSize(size);
+  const parsed = validateSizeFormat(normalized);
+  const totalPixels = parsed.width * parsed.height;
+  if (totalPixels < MIN_WAN27_TOTAL_PIXELS || totalPixels > maxPixels) {
+    const limit = maxPixels === MAX_WAN27_PRO_T2I_PIXELS ? "4096*4096" : "2048*2048";
+    throw new Error(
+      `DashScope ${model} requires total pixels between 768*768 and ${limit} ` +
+      `for the current request. Received ${normalized} (${totalPixels} pixels).`
+    );
+  }
+  const ratio = parsed.width / parsed.height;
+  if (ratio < 1 / 8 || ratio > 8) {
+    throw new Error(
+      `DashScope wan2.7 image models support aspect ratios in [1:8, 8:1]. ` +
+      `Received ${normalized} (ratio ${ratio.toFixed(3)}).`
+    );
+  }
+  return normalized;
+}
+
 function getQwenFixedSizeFromAspectRatio(ar: string | null, quality: CliArgs["quality"]): string {
  if (quality === "normal") {
    console.warn(
@@ -331,9 +457,16 @@ function validateQwenFixedSize(size: string): string {

 export function resolveSizeForModel(
  model: string,
-  args: Pick<CliArgs, "size" | "aspectRatio" | "quality">,
+  args: Pick<CliArgs, "size" | "aspectRatio" | "quality"> & { referenceImages?: string[] },
 ): string {
  const spec = getModelSpec(model);
+  const referenceCount = args.referenceImages?.length ?? 0;
+
+  if (spec.family === "wan27") {
+    const maxPixels = getWan27MaxPixels(model, referenceCount > 0);
+    if (args.size) return validateWan27Size(args.size, maxPixels, model);
+    return getWan27SizeFromAspectRatio(args.aspectRatio, args.quality, maxPixels);
+  }

  if (args.size) {
    if (spec.family === "qwen2") return validateQwen2Size(args.size);
@@ -357,6 +490,14 @@ function buildParameters(
  family: DashScopeModelFamily,
  size: string,
 ): Record<string, unknown> {
+  if (family === "wan27") {
+    return {
+      size,
+      n: 1,
+      watermark: false,
+    };
+  }
+
  const parameters: Record<string, unknown> = {
    prompt_extend: false,
    size,
@@ -419,23 +560,44 @@ export async function generateImage(
  const apiKey = getApiKey();
  if (!apiKey) throw new Error("DASHSCOPE_API_KEY is required");

-  if (args.referenceImages.length > 0) {
+  const spec = getModelSpec(model);
+
+  if (args.referenceImages.length > 0 && spec.family !== "wan27") {
    throw new Error(
-      "Reference images are not supported with DashScope provider in baoyu-image-gen. Use --provider google with a Gemini multimodal model."
+      "Reference images are not supported with this DashScope model. Use a wan2.7 image model (--model wan2.7-image-pro or wan2.7-image), or switch to --provider google with a Gemini multimodal model."
+    );
+  }
+
+  if (args.referenceImages.length > WAN27_MAX_REFERENCE_IMAGES) {
+    throw new Error(
+      `DashScope wan2.7 image models accept at most ${WAN27_MAX_REFERENCE_IMAGES} reference images. Received ${args.referenceImages.length}.`
+    );
+  }
+
+  if (spec.family === "wan27" && args.n !== 1) {
+    throw new Error(
+      "DashScope wan2.7 image models in baoyu-image-gen support exactly one output image per request (extra images would be billed but discarded). Remove --n or use --n 1."
    );
  }

-  const spec = getModelSpec(model);
  const size = resolveSizeForModel(model, args);
  const url = `${getBaseUrl()}/api/v1/services/aigc/multimodal-generation/generation`;

+  const content: Array<Record<string, unknown>> = [];
+  if (spec.family === "wan27" && args.referenceImages.length > 0) {
+    for (const refPath of args.referenceImages) {
+      content.push({ image: await loadReferenceImage(refPath) });
+    }
+  }
+  content.push({ text: prompt });
+
  const body = {
    model,
    input: {
      messages: [
        {
          role: "user",
-          content: [{ text: prompt }],
+          content,
        },
      ],
    },
@@ -50,6 +50,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -15,6 +15,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -50,6 +50,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -1,14 +1,47 @@
 import assert from "node:assert/strict";
 import test from "node:test";

+import type { CliArgs } from "../types.ts";
 import {
+  buildOpenAIGenerationsBody,
  extractImageFromResponse,
+  getDefaultModel,
+  getOpenAIAspectRatio,
+  getOpenAIImageApiDialect,
+  getOpenAIResolution,
  getMimeType,
  getOpenAISize,
+  getOrientationFromAspectRatio,
+  inferAspectRatioFromSize,
+  inferResolutionFromSize,
  parseAspectRatio,
+  validateArgs,
 } from "./openai.ts";

+function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
+  return {
+    prompt: null,
+    promptFiles: [],
+    imagePath: null,
+    provider: null,
+    model: null,
+    aspectRatio: null,
+    size: null,
+    quality: "2k",
+    imageSize: null,
+    imageApiDialect: null,
+    referenceImages: [],
+    n: 1,
+    batchFile: null,
+    jobs: null,
+    json: false,
+    help: false,
+    ...overrides,
+  };
+}
+
 test("OpenAI aspect-ratio parsing and size selection match model families", () => {
+  assert.equal(getDefaultModel(), "gpt-image-2");
  assert.deepEqual(parseAspectRatio("16:9"), { width: 16, height: 9 });
  assert.equal(parseAspectRatio("wide"), null);
  assert.equal(parseAspectRatio("0:1"), null);
@@ -18,6 +51,96 @@ test("OpenAI aspect-ratio parsing and size selection match model families", () =
  assert.equal(getOpenAISize("dall-e-2", "16:9", "2k"), "1024x1024");
  assert.equal(getOpenAISize("gpt-image-1.5", "16:9", "2k"), "1536x1024");
  assert.equal(getOpenAISize("gpt-image-1.5", "4:3", "2k"), "1024x1024");
+  assert.equal(getOpenAISize("gpt-image-2", "16:9", "2k"), "2048x1152");
+  assert.equal(getOpenAISize("gpt-image-2", "9:16", "2k"), "1152x2048");
+  assert.equal(getOpenAISize("gpt-image-2", "4:3", "2k"), "2048x1536");
+  assert.equal(getOpenAISize("gpt-image-2", "2.35:1", "normal"), "1248x528");
+  assert.equal(inferAspectRatioFromSize("1536x1024"), "3:2");
+  assert.equal(inferResolutionFromSize("1536x1024"), "2K");
+  assert.equal(getOpenAIAspectRatio({ aspectRatio: null, size: "2048x1152" }), "16:9");
+  assert.equal(getOpenAIResolution({ imageSize: null, size: "2048x1152", quality: "normal" }), "2K");
+  assert.equal(getOrientationFromAspectRatio("16:9"), "landscape");
+  assert.equal(getOrientationFromAspectRatio("9:16"), "portrait");
+  assert.equal(getOrientationFromAspectRatio("1:1"), null);
+  assert.equal(getOpenAIImageApiDialect({ imageApiDialect: null }), "openai-native");
+});
+
+test("OpenAI generations body switches between native and ratio-metadata dialects", () => {
+  assert.deepEqual(
+    buildOpenAIGenerationsBody("Draw a skyline", "gpt-image-2", {
+      aspectRatio: "16:9",
+      size: null,
+      quality: "2k",
+      imageSize: null,
+      imageApiDialect: null,
+    }),
+    {
+      model: "gpt-image-2",
+      prompt: "Draw a skyline",
+      size: "2048x1152",
+      quality: "high",
+    },
+  );
+
+  assert.deepEqual(
+    buildOpenAIGenerationsBody("Draw a skyline", "gemini-3-pro-image-preview", {
+      aspectRatio: "16:9",
+      size: null,
+      quality: "2k",
+      imageSize: null,
+      imageApiDialect: "ratio-metadata",
+    }),
+    {
+      model: "gemini-3-pro-image-preview",
+      prompt: "Draw a skyline",
+      size: "16:9",
+      metadata: {
+        resolution: "2K",
+        orientation: "landscape",
+      },
+    },
+  );
+
+  assert.deepEqual(
+    buildOpenAIGenerationsBody("Draw a portrait", "gemini-3-pro-image-preview", {
+      aspectRatio: null,
+      size: "1152x2048",
+      quality: "normal",
+      imageSize: null,
+      imageApiDialect: "ratio-metadata",
+    }),
+    {
+      model: "gemini-3-pro-image-preview",
+      prompt: "Draw a portrait",
+      size: "9:16",
+      metadata: {
+        resolution: "2K",
+        orientation: "portrait",
+      },
+    },
+  );
+});
+
+test("OpenAI validates gpt-image-2 custom size constraints", () => {
+  assert.doesNotThrow(() =>
+    validateArgs("gpt-image-2", makeArgs({ size: "3840x2160" })),
+  );
+  assert.doesNotThrow(() =>
+    validateArgs("gpt-image-2-2026-04-21", makeArgs({ aspectRatio: "2.35:1" })),
+  );
+
+  assert.throws(
+    () => validateArgs("gpt-image-2", makeArgs({ size: "1024x576" })),
+    /total pixels/,
+  );
+  assert.throws(
+    () => validateArgs("gpt-image-2", makeArgs({ size: "1025x1024" })),
+    /multiples of 16px/,
+  );
+  assert.throws(
+    () => validateArgs("gpt-image-2", makeArgs({ aspectRatio: "4:1" })),
+    /must not exceed 3:1/,
+  );
 });

 test("OpenAI mime-type detection covers supported reference image extensions", () => {
@@ -1,9 +1,9 @@
 import path from "node:path";
 import { readFile } from "node:fs/promises";
-import type { CliArgs } from "../types";
+import type { CliArgs, OpenAIImageApiDialect } from "../types";

 export function getDefaultModel(): string {
-  return process.env.OPENAI_IMAGE_MODEL || "gpt-image-1.5";
+  return process.env.OPENAI_IMAGE_MODEL || "gpt-image-2";
 }

 type OpenAIImageResponse = { data: Array<{ url?: string; b64_json?: string }> };
@@ -23,6 +23,57 @@ type SizeMapping = {
  portrait: string;
 };

+type OpenAIGenerationsBody = Record<string, unknown>;
+
+function isGptImageModel(model: string): boolean {
+  return model.includes("gpt-image");
+}
+
+function isGptImage2Model(model: string): boolean {
+  return model.includes("gpt-image-2");
+}
+
+function roundToMultiple(value: number, multiple: number): number {
+  return Math.max(multiple, Math.round(value / multiple) * multiple);
+}
+
+function buildGptImage2SizeFromAspectRatio(
+  ar: string | null,
+  quality: CliArgs["quality"],
+): string {
+  const parsed = ar ? parseAspectRatio(ar) : null;
+  const ratio = parsed ? parsed.width / parsed.height : 1;
+
+  if (!parsed || Math.abs(ratio - 1) < 0.1) {
+    const edge = quality === "2k" ? 2048 : 1024;
+    return `${edge}x${edge}`;
+  }
+
+  const targetLongEdge = quality === "2k" ? 2048 : 1024;
+  let width: number;
+  let height: number;
+
+  if (ratio > 1) {
+    width = targetLongEdge;
+    height = roundToMultiple(width / ratio, 16);
+  } else {
+    height = targetLongEdge;
+    width = roundToMultiple(height * ratio, 16);
+  }
+
+  while (width * height < 655_360) {
+    if (ratio > 1) {
+      width += 16;
+      height = roundToMultiple(width / ratio, 16);
+    } else {
+      height += 16;
+      width = roundToMultiple(height * ratio, 16);
+    }
+  }
+
+  return `${width}x${height}`;
+}
+
 export function getOpenAISize(
  model: string,
  ar: string | null,
@@ -35,6 +86,10 @@ export function getOpenAISize(
    return "1024x1024";
  }

+  if (isGptImage2Model(model)) {
+    return buildGptImage2SizeFromAspectRatio(ar, quality);
+  }
+
  const sizes: SizeMapping = isDalle3
    ? {
        square: "1024x1024",
@@ -60,6 +115,166 @@ export function getOpenAISize(
  return sizes.square;
 }

+function parsePixelSize(value: string): { width: number; height: number } | null {
+  const match = value.match(/^(\d+)\s*[xX]\s*(\d+)$/);
+  if (!match) return null;
+
+  const width = parseInt(match[1]!, 10);
+  const height = parseInt(match[2]!, 10);
+  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) {
+    return null;
+  }
+
+  return { width, height };
+}
+
+function gcd(a: number, b: number): number {
+  let x = Math.abs(a);
+  let y = Math.abs(b);
+  while (y !== 0) {
+    const next = x % y;
+    x = y;
+    y = next;
+  }
+  return x || 1;
+}
+
+export function getOpenAIImageApiDialect(args: Pick<CliArgs, "imageApiDialect">): OpenAIImageApiDialect {
+  return args.imageApiDialect ?? "openai-native";
+}
+
+export function inferAspectRatioFromSize(size: string | null): string | null {
+  if (!size) return null;
+  const parsed = parsePixelSize(size);
+  if (!parsed) return null;
+
+  const divisor = gcd(parsed.width, parsed.height);
+  return `${parsed.width / divisor}:${parsed.height / divisor}`;
+}
+
+export function inferResolutionFromSize(size: string | null): "1K" | "2K" | "4K" | null {
+  if (!size) return null;
+  const parsed = parsePixelSize(size);
+  if (!parsed) return null;
+
+  const longestEdge = Math.max(parsed.width, parsed.height);
+  if (longestEdge <= 1024) return "1K";
+  if (longestEdge <= 2048) return "2K";
+  return "4K";
+}
+
+export function getOpenAIAspectRatio(args: Pick<CliArgs, "aspectRatio" | "size">): string {
+  return args.aspectRatio ?? inferAspectRatioFromSize(args.size) ?? "1:1";
+}
+
+export function getOpenAIResolution(
+  args: Pick<CliArgs, "imageSize" | "size" | "quality">
+): "1K" | "2K" | "4K" {
+  if (args.imageSize === "1K" || args.imageSize === "2K" || args.imageSize === "4K") {
+    return args.imageSize;
+  }
+
+  const inferred = inferResolutionFromSize(args.size);
+  if (inferred) return inferred;
+
+  return args.quality === "normal" ? "1K" : "2K";
+}
+
+function getOpenAIQuality(model: string, quality: CliArgs["quality"]): "standard" | "hd" | "medium" | "high" | null {
+  if (model.includes("dall-e-3")) {
+    return quality === "2k" ? "hd" : "standard";
+  }
+
+  if (isGptImageModel(model)) {
+    return quality === "2k" ? "high" : "medium";
+  }
+
+  return null;
+}
+
+export function getOrientationFromAspectRatio(ar: string): "landscape" | "portrait" | null {
+  const parsed = parseAspectRatio(ar);
+  if (!parsed) return null;
+
+  const ratio = parsed.width / parsed.height;
+  if (Math.abs(ratio - 1) < 0.1) return null;
+  return ratio > 1 ? "landscape" : "portrait";
+}
+
+export function buildOpenAIGenerationsBody(
+  prompt: string,
+  model: string,
+  args: Pick<CliArgs, "aspectRatio" | "size" | "quality" | "imageSize" | "imageApiDialect">
+): OpenAIGenerationsBody {
+  if (getOpenAIImageApiDialect(args) === "ratio-metadata") {
+    const aspectRatio = getOpenAIAspectRatio(args);
+    const metadata: Record<string, string> = {
+      resolution: getOpenAIResolution(args),
+    };
+    const orientation = getOrientationFromAspectRatio(aspectRatio);
+    if (orientation) metadata.orientation = orientation;
+
+    return {
+      model,
+      prompt,
+      size: aspectRatio,
+      metadata,
+    };
+  }
+
+  const body: OpenAIGenerationsBody = {
+    model,
+    prompt,
+    size: args.size || getOpenAISize(model, args.aspectRatio, args.quality),
+  };
+
+  const quality = getOpenAIQuality(model, args.quality);
+  if (quality) {
+    body.quality = quality;
+  }
+
+  return body;
+}
+
+export function validateArgs(model: string, args: CliArgs): void {
+  if (!isGptImage2Model(model)) return;
+
+  if (args.aspectRatio && !args.size) {
+    const parsed = parseAspectRatio(args.aspectRatio);
+    if (!parsed) {
+      throw new Error(`Invalid gpt-image-2 aspect ratio: ${args.aspectRatio}`);
+    }
+    const ratio = parsed.width / parsed.height;
+    if (Math.max(ratio, 1 / ratio) > 3) {
+      throw new Error("gpt-image-2 aspect ratio must not exceed 3:1.");
+    }
+  }
+
+  if (!args.size) return;
+
+  const parsedSize = parsePixelSize(args.size);
+  if (!parsedSize) {
+    throw new Error(`Invalid gpt-image-2 --size: ${args.size}. Expected <width>x<height>.`);
+  }
+
+  const { width, height } = parsedSize;
+  const totalPixels = width * height;
+  const ratio = Math.max(width, height) / Math.min(width, height);
+
+  if (Math.max(width, height) > 3840) {
+    throw new Error("gpt-image-2 --size maximum edge length must be 3840px or less.");
+  }
+  if (width % 16 !== 0 || height % 16 !== 0) {
+    throw new Error("gpt-image-2 --size width and height must both be multiples of 16px.");
+  }
+  if (ratio > 3) {
+    throw new Error("gpt-image-2 --size long edge to short edge ratio must not exceed 3:1.");
+  }
+  if (totalPixels < 655_360 || totalPixels > 8_294_400) {
+    throw new Error("gpt-image-2 --size total pixels must be between 655,360 and 8,294,400.");
+  }
+}
+
 export async function generateImage(
  prompt: string,
  model: string,
@@ -78,18 +293,28 @@ export async function generateImage(
    return generateWithChatCompletions(baseURL, apiKey, prompt, model);
  }

-  const size = args.size || getOpenAISize(model, args.aspectRatio, args.quality);
+  const imageApiDialect = getOpenAIImageApiDialect(args);

  if (args.referenceImages.length > 0) {
-    if (model.includes("dall-e-2") || model.includes("dall-e-3")) {
+    if (imageApiDialect !== "openai-native") {
      throw new Error(
-        "Reference images with OpenAI in this skill require GPT Image models. Use --model gpt-image-1.5 (or another gpt-image model)."
+        "Reference images are not supported with the ratio-metadata OpenAI dialect yet. Use openai-native, Google, Azure, OpenRouter, MiniMax, Seedream, or Replicate for image-edit workflows."
      );
    }
+    if (model.includes("dall-e-2") || model.includes("dall-e-3")) {
+      throw new Error(
+        "Reference images with OpenAI in this skill require GPT Image models. Use --model gpt-image-2 (or another gpt-image model)."
+      );
+    }
+    const size = args.size || getOpenAISize(model, args.aspectRatio, args.quality);
    return generateWithOpenAIEdits(baseURL, apiKey, prompt, model, size, args.referenceImages, args.quality);
  }

-  return generateWithOpenAIGenerations(baseURL, apiKey, prompt, model, size, args.quality);
+  return generateWithOpenAIGenerations(
+    baseURL,
+    apiKey,
+    buildOpenAIGenerationsBody(prompt, model, args)
+  );
 }

 async function generateWithChatCompletions(
@@ -129,17 +354,8 @@ async function generateWithChatCompletions(
 async function generateWithOpenAIGenerations(
  baseURL: string,
  apiKey: string,
-  prompt: string,
-  model: string,
-  size: string,
-  quality: CliArgs["quality"]
+  body: OpenAIGenerationsBody
 ): Promise<Uint8Array> {
-  const body: Record<string, any> = { model, prompt, size };
-
-  if (model.includes("dall-e-3")) {
-    body.quality = quality === "2k" ? "hd" : "standard";
-  }
-
  const res = await fetch(`${baseURL}/images/generations`, {
    method: "POST",
    headers: {
@@ -172,8 +388,9 @@ async function generateWithOpenAIEdits(
  form.append("prompt", prompt);
  form.append("size", size);

-  if (model.includes("gpt-image")) {
-    form.append("quality", quality === "2k" ? "high" : "medium");
+  const openAIQuality = getOpenAIQuality(model, quality);
+  if (openAIQuality && openAIQuality !== "standard" && openAIQuality !== "hd") {
+    form.append("quality", openAIQuality);
  }

  for (const refPath of referenceImages) {
@@ -28,6 +28,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -5,7 +5,10 @@ import type { CliArgs } from "../types.ts";
 import {
  buildInput,
  extractOutputUrl,
+  getDefaultModel,
+  getModelFamily,
  parseModelId,
+  validateArgs,
 } from "./replicate.ts";

 function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
@@ -16,9 +19,12 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    provider: null,
    model: null,
    aspectRatio: null,
+    aspectRatioSource: null,
    size: null,
    quality: null,
    imageSize: null,
+    imageSizeSource: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -29,10 +35,24 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
  };
 }

-test("Replicate model parsing accepts official formats and rejects malformed ones", () => {
-  assert.deepEqual(parseModelId("google/nano-banana-pro"), {
+test("Replicate default model now points at nano-banana-2", () => {
+  const previous = process.env.REPLICATE_IMAGE_MODEL;
+  delete process.env.REPLICATE_IMAGE_MODEL;
+  try {
+    assert.equal(getDefaultModel(), "google/nano-banana-2");
+  } finally {
+    if (previous == null) {
+      delete process.env.REPLICATE_IMAGE_MODEL;
+    } else {
+      process.env.REPLICATE_IMAGE_MODEL = previous;
+    }
+  }
+});
+
+test("Replicate model parsing and family detection accept supported official ids", () => {
+  assert.deepEqual(parseModelId("google/nano-banana-2"), {
    owner: "google",
-    name: "nano-banana-pro",
+    name: "nano-banana-2",
    version: null,
  });
  assert.deepEqual(parseModelId("owner/model:abc123"), {
@@ -41,46 +61,224 @@ test("Replicate model parsing accepts official formats and rejects malformed one
    version: "abc123",
  });

+  assert.equal(getModelFamily("google/nano-banana-pro"), "nano-banana");
+  assert.equal(getModelFamily("bytedance/seedream-4.5"), "seedream45");
+  assert.equal(getModelFamily("bytedance/seedream-5-lite"), "seedream5lite");
+  assert.equal(getModelFamily("wan-video/wan-2.7-image"), "wan27image");
+  assert.equal(getModelFamily("wan-video/wan-2.7-image-pro"), "wan27imagepro");
+  assert.equal(getModelFamily("stability-ai/sdxl"), "unknown");
+
  assert.throws(
    () => parseModelId("just-a-model-name"),
    /Invalid Replicate model format/,
  );
 });

-test("Replicate input builder maps aspect ratio, image count, quality, and refs", () => {
+test("Replicate nano-banana input builder maps refs, aspect ratio, and quality presets", () => {
  assert.deepEqual(
    buildInput(
+      "google/nano-banana-2",
      "A robot painter",
      makeArgs({
        aspectRatio: "16:9",
        quality: "2k",
-        n: 3,
      }),
      ["data:image/png;base64,AAAA"],
    ),
    {
      prompt: "A robot painter",
-      aspect_ratio: "16:9",
-      number_of_images: 3,
      resolution: "2K",
      output_format: "png",
+      aspect_ratio: "16:9",
      image_input: ["data:image/png;base64,AAAA"],
    },
  );

  assert.deepEqual(
-    buildInput("A robot painter", makeArgs({ quality: "normal" }), ["ref"]),
+    buildInput(
+      "google/nano-banana-2",
+      "A robot painter",
+      makeArgs({ size: "1024x1024", quality: "normal" }),
+      [],
+    ),
    {
      prompt: "A robot painter",
-      aspect_ratio: "match_input_image",
      resolution: "1K",
      output_format: "png",
-      image_input: ["ref"],
+      aspect_ratio: "1:1",
    },
  );
 });

-test("Replicate output extraction supports string, array, and object URLs", () => {
+test("Replicate Seedream and Wan inputs use family-specific request fields", () => {
+  assert.deepEqual(
+    buildInput(
+      "bytedance/seedream-4.5",
+      "A cinematic portrait",
+      makeArgs({ quality: "2k", referenceImages: ["local.png"] }),
+      ["data:image/png;base64,AAAA"],
+    ),
+    {
+      prompt: "A cinematic portrait",
+      size: "4K",
+      image_input: ["data:image/png;base64,AAAA"],
+      aspect_ratio: "match_input_image",
+    },
+  );
+
+  assert.deepEqual(
+    buildInput(
+      "bytedance/seedream-4.5",
+      "A cinematic portrait",
+      makeArgs({ size: "1536x1024" }),
+      [],
+    ),
+    {
+      prompt: "A cinematic portrait",
+      size: "custom",
+      width: 1536,
+      height: 1024,
+    },
+  );
+
+  assert.deepEqual(
+    buildInput(
+      "bytedance/seedream-5-lite",
+      "A poster",
+      makeArgs({ aspectRatio: "21:9", quality: "2k" }),
+      [],
+    ),
+    {
+      prompt: "A poster",
+      size: "3K",
+      aspect_ratio: "21:9",
+    },
+  );
+
+  assert.deepEqual(
+    buildInput(
+      "wan-video/wan-2.7-image",
+      "A storyboard frame",
+      makeArgs({ aspectRatio: "16:9", quality: "2k" }),
+      [],
+    ),
+    {
+      prompt: "A storyboard frame",
+      size: "2048*1152",
+    },
+  );
+
+  assert.deepEqual(
+    buildInput(
+      "wan-video/wan-2.7-image-pro",
+      "Blend these references",
+      makeArgs({ size: "2K", referenceImages: ["a.png", "b.png"] }),
+      ["ref-a", "ref-b"],
+    ),
+    {
+      prompt: "Blend these references",
+      size: "2K",
+      images: ["ref-a", "ref-b"],
+    },
+  );
+});
+
+test("Replicate validateArgs blocks misleading multi-output and unsupported family options locally", () => {
+  assert.throws(
+    () =>
+      validateArgs(
+        "google/nano-banana-2",
+        makeArgs({ n: 2 }),
+      ),
+    /exactly one output image/,
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "bytedance/seedream-4.5",
+        makeArgs({ size: "1K" }),
+      ),
+    /2K, 4K, or an explicit WxH size/,
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "bytedance/seedream-5-lite",
+        makeArgs({ size: "4K" }),
+      ),
+    /supports 2K or 3K output/,
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "wan-video/wan-2.7-image",
+        makeArgs({ referenceImages: new Array(10).fill("ref.png") }),
+      ),
+    /at most 9 reference images/,
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "wan-video/wan-2.7-image-pro",
+        makeArgs({ referenceImages: ["ref.png"], size: "4K" }),
+      ),
+    /only supports 4K text-to-image/,
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "stability-ai/sdxl",
+        makeArgs({ aspectRatio: "16:9" }),
+      ),
+    /compatibility list/,
+  );
+
+  assert.doesNotThrow(() =>
+    validateArgs(
+      "google/nano-banana-2",
+      makeArgs({ imageSize: "2K", imageSizeSource: "config" }),
+    ),
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "google/nano-banana-2",
+        makeArgs({ imageSize: "2K", imageSizeSource: "cli" }),
+      ),
+    /do not use --imageSize/,
+  );
+
+  assert.doesNotThrow(() =>
+    validateArgs(
+      "stability-ai/sdxl",
+      makeArgs({ aspectRatio: "16:9", aspectRatioSource: "config" }),
+    ),
+  );
+
+  assert.throws(
+    () =>
+      validateArgs(
+        "stability-ai/sdxl",
+        makeArgs({ aspectRatio: "16:9", aspectRatioSource: "cli" }),
+      ),
+    /compatibility list/,
+  );
+
+  assert.doesNotThrow(() =>
+    validateArgs(
+      "stability-ai/sdxl",
+      makeArgs(),
+    ),
+  );
+});
+
+test("Replicate output extraction supports single outputs and rejects silent multi-image drops", () => {
  assert.equal(
    extractOutputUrl({ output: "https://example.com/a.png" } as never),
    "https://example.com/a.png",
@@ -94,6 +292,17 @@ test("Replicate output extraction supports string, array, and object URLs", () =
    "https://example.com/c.png",
  );

+  assert.throws(
+    () =>
+      extractOutputUrl({
+        output: [
+          "https://example.com/one.png",
+          "https://example.com/two.png",
+        ],
+      } as never),
+    /supports saving exactly one image/,
+  );
+
  assert.throws(
    () => extractOutputUrl({ output: { invalid: true } } as never),
    /Unexpected Replicate output format/,
@@ -2,10 +2,37 @@ import path from "node:path";
 import { readFile } from "node:fs/promises";
 import type { CliArgs } from "../types";

-const DEFAULT_MODEL = "google/nano-banana-pro";
+const DEFAULT_MODEL = "google/nano-banana-2";
 const SYNC_WAIT_SECONDS = 60;
 const POLL_INTERVAL_MS = 2000;
 const MAX_POLL_MS = 300_000;
+const DOCUMENTED_REPLICATE_ASPECT_RATIOS = new Set([
+  "1:1",
+  "2:3",
+  "3:2",
+  "3:4",
+  "4:3",
+  "5:4",
+  "4:5",
+  "9:16",
+  "16:9",
+  "21:9",
+]);
+
+export type ReplicateModelFamily =
+  | "nano-banana"
+  | "seedream45"
+  | "seedream5lite"
+  | "wan27image"
+  | "wan27imagepro"
+  | "unknown";
+
+type PixelSize = {
+  width: number;
+  height: number;
+};
+
+type Seedream45Size = "2K" | "4K" | { width: number; height: number };

 export function getDefaultModel(): string {
  return process.env.REPLICATE_IMAGE_MODEL || DEFAULT_MODEL;
@@ -20,6 +47,40 @@ function getBaseUrl(): string {
  return base.replace(/\/+$/g, "");
 }

+function normalizeModelId(model: string): string {
+  return model.trim().toLowerCase().split(":")[0]!;
+}
+
+export function getModelFamily(model: string): ReplicateModelFamily {
+  const normalized = normalizeModelId(model);
+
+  if (
+    normalized === "google/nano-banana" ||
+    normalized === "google/nano-banana-pro" ||
+    normalized === "google/nano-banana-2"
+  ) {
+    return "nano-banana";
+  }
+
+  if (normalized === "bytedance/seedream-4.5") {
+    return "seedream45";
+  }
+
+  if (normalized === "bytedance/seedream-5-lite") {
+    return "seedream5lite";
+  }
+
+  if (normalized === "wan-video/wan-2.7-image") {
+    return "wan27image";
+  }
+
+  if (normalized === "wan-video/wan-2.7-image-pro") {
+    return "wan27imagepro";
+  }
+
+  return "unknown";
+}
+
 export function parseModelId(model: string): { owner: string; name: string; version: string | null } {
  const [ownerName, version] = model.split(":");
  const parts = ownerName!.split("/");
@@ -31,27 +92,219 @@ export function parseModelId(model: string): { owner: string; name: string; vers
  return { owner: parts[0], name: parts[1], version: version || null };
 }

-export function buildInput(prompt: string, args: CliArgs, referenceImages: string[]): Record<string, unknown> {
-  const input: Record<string, unknown> = { prompt };
+function parsePixelSize(value: string): PixelSize | null {
+  const match = value.trim().match(/^(\d+)\s*[xX*]\s*(\d+)$/);
+  if (!match) return null;
+
+  const width = parseInt(match[1]!, 10);
+  const height = parseInt(match[2]!, 10);
+  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) {
+    return null;
+  }
+
+  return { width, height };
+}
+
+function parseAspectRatio(value: string): PixelSize | null {
+  const match = value.trim().match(/^(\d+)\s*:\s*(\d+)$/);
+  if (!match) return null;
+
+  const width = parseInt(match[1]!, 10);
+  const height = parseInt(match[2]!, 10);
+  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) {
+    return null;
+  }
+
+  return { width, height };
+}
+
+function gcd(a: number, b: number): number {
+  let x = Math.abs(a);
+  let y = Math.abs(b);
+
+  while (y !== 0) {
+    const next = x % y;
+    x = y;
+    y = next;
+  }
+
+  return x || 1;
+}
+
+function inferAspectRatioFromSize(size: string): string | null {
+  const parsed = parsePixelSize(size);
+  if (!parsed) return null;
+
+  const divisor = gcd(parsed.width, parsed.height);
+  const normalized = `${parsed.width / divisor}:${parsed.height / divisor}`;
+  if (!DOCUMENTED_REPLICATE_ASPECT_RATIOS.has(normalized)) {
+    return null;
+  }
+
+  return normalized;
+}
+
+function getQualityPreset(args: CliArgs): "normal" | "2k" {
+  return args.quality === "normal" ? "normal" : "2k";
+}
+
+function validateDocumentedAspectRatio(model: string, aspectRatio: string): void {
+  if (aspectRatio === "match_input_image") {
+    return;
+  }
+
+  if (DOCUMENTED_REPLICATE_ASPECT_RATIOS.has(aspectRatio)) {
+    return;
+  }
+
+  throw new Error(
+    `Replicate model ${model} does not support aspect ratio ${aspectRatio}. Supported values: ${Array.from(DOCUMENTED_REPLICATE_ASPECT_RATIOS).join(", ")}`
+  );
+}
+
+function getRequestedAspectRatio(model: string, args: CliArgs): string | null {
+  if (args.aspectRatio) {
+    validateDocumentedAspectRatio(model, args.aspectRatio);
+    return args.aspectRatio;
+  }
+
+  if (!args.size) return null;
+
+  const inferred = inferAspectRatioFromSize(args.size);
+  if (!inferred) {
+    throw new Error(
+      `Replicate model ${model} cannot derive a supported aspect ratio from --size ${args.size}. Use one of: ${Array.from(DOCUMENTED_REPLICATE_ASPECT_RATIOS).join(", ")}`
+    );
+  }
+
+  return inferred;
+}
+
+function getNanoBananaResolution(args: CliArgs): "1K" | "2K" {
+  if (args.size) {
+    const parsed = parsePixelSize(args.size);
+    if (!parsed) {
+      throw new Error("Replicate nano-banana --size must be in WxH format, for example 1536x1024.");
+    }
+
+    const longestEdge = Math.max(parsed.width, parsed.height);
+    if (longestEdge <= 1024) return "1K";
+    if (longestEdge <= 2048) return "2K";
+    throw new Error("Replicate nano-banana only supports sizes that map to 1K or 2K output.");
+  }
+
+  return getQualityPreset(args) === "normal" ? "1K" : "2K";
+}
+
+function resolveSeedream45Size(args: CliArgs): Seedream45Size {
+  if (args.size) {
+    const upper = args.size.trim().toUpperCase();
+    if (upper === "2K" || upper === "4K") {
+      return upper;
+    }
+
+    const parsed = parsePixelSize(args.size);
+    if (!parsed) {
+      throw new Error("Replicate Seedream 4.5 --size must be 2K, 4K, or an explicit WxH size.");
+    }
+    if (parsed.width < 1024 || parsed.width > 4096 || parsed.height < 1024 || parsed.height > 4096) {
+      throw new Error("Replicate Seedream 4.5 custom --size must keep width and height between 1024 and 4096.");
+    }
+    return parsed;
+  }
+
+  return getQualityPreset(args) === "normal" ? "2K" : "4K";
+}
+
+function resolveSeedream5LiteSize(args: CliArgs): "2K" | "3K" {
+  if (args.size) {
+    const upper = args.size.trim().toUpperCase();
+    if (upper === "2K" || upper === "3K") {
+      return upper;
+    }
+
+    throw new Error("Replicate Seedream 5 Lite currently supports 2K or 3K output in this tool.");
+  }
+
+  return getQualityPreset(args) === "normal" ? "2K" : "3K";
+}
+
+function formatCustomWanSize(size: PixelSize): string {
+  return `${size.width}*${size.height}`;
+}
+
+function resolveWanSizeFromAspectRatio(
+  aspectRatio: string,
+  maxDimension: number,
+): string {
+  const parsedRatio = parseAspectRatio(aspectRatio);
+  if (!parsedRatio) {
+    throw new Error(`Replicate Wan aspect ratio must be in W:H format, got ${aspectRatio}.`);
+  }
+
+  const scale = Math.min(maxDimension / parsedRatio.width, maxDimension / parsedRatio.height);
+  const width = Math.max(1, Math.floor(parsedRatio.width * scale));
+  const height = Math.max(1, Math.floor(parsedRatio.height * scale));
+  return formatCustomWanSize({ width, height });
+}
+
+function resolveWanSize(family: "wan27image" | "wan27imagepro", args: CliArgs): "1K" | "2K" | "4K" | string {
+  const referenceMode = args.referenceImages.length > 0;
+  const maxDimension = family === "wan27imagepro" && !referenceMode ? 4096 : 2048;
+
+  if (args.size) {
+    const upper = args.size.trim().toUpperCase();
+    if (upper === "1K" || upper === "2K" || upper === "4K") {
+      if (upper === "4K" && family !== "wan27imagepro") {
+        throw new Error("Replicate Wan 2.7 Image only supports 1K, 2K, or custom sizes up to 2048px.");
+      }
+      if (upper === "4K" && referenceMode) {
+        throw new Error("Replicate Wan 2.7 Image Pro only supports 4K text-to-image. Remove --ref or lower the size.");
+      }
+      return upper;
+    }
+
+    const parsed = parsePixelSize(args.size);
+    if (!parsed) {
+      throw new Error("Replicate Wan --size must be 1K, 2K, 4K, or an explicit WxH size.");
+    }
+    if (parsed.width > maxDimension || parsed.height > maxDimension) {
+      throw new Error(
+        `Replicate ${family === "wan27imagepro" ? "Wan 2.7 Image Pro" : "Wan 2.7 Image"} custom --size must keep width and height at or below ${maxDimension}px in the current mode.`
+      );
+    }
+    return formatCustomWanSize(parsed);
+  }

  if (args.aspectRatio) {
-    input.aspect_ratio = args.aspectRatio;
+    return resolveWanSizeFromAspectRatio(
+      args.aspectRatio,
+      getQualityPreset(args) === "normal" ? 1024 : 2048,
+    );
+  }
+
+  return getQualityPreset(args) === "normal" ? "1K" : "2K";
+}
+
+function buildNanoBananaInput(
+  prompt: string,
+  model: string,
+  args: CliArgs,
+  referenceImages: string[],
+): Record<string, unknown> {
+  const input: Record<string, unknown> = {
+    prompt,
+    resolution: getNanoBananaResolution(args),
+    output_format: "png",
+  };
+
+  const aspectRatio = getRequestedAspectRatio(model, args);
+  if (aspectRatio) {
+    input.aspect_ratio = aspectRatio;
  } else if (referenceImages.length > 0) {
    input.aspect_ratio = "match_input_image";
  }

-  if (args.n > 1) {
-    input.number_of_images = args.n;
-  }
-
-  if (args.quality === "normal") {
-    input.resolution = "1K";
-  } else if (args.quality === "2k") {
-    input.resolution = "2K";
-  }
-
-  input.output_format = "png";
-
  if (referenceImages.length > 0) {
    input.image_input = referenceImages;
  }
@@ -59,6 +312,158 @@ export function buildInput(prompt: string, args: CliArgs, referenceImages: strin
  return input;
 }

+function buildSeedreamInput(
+  family: "seedream45" | "seedream5lite",
+  prompt: string,
+  model: string,
+  args: CliArgs,
+  referenceImages: string[],
+): Record<string, unknown> {
+  const size = family === "seedream45" ? resolveSeedream45Size(args) : resolveSeedream5LiteSize(args);
+  const input: Record<string, unknown> = {
+    prompt,
+  };
+
+  if (family === "seedream45" && typeof size === "object") {
+    input.size = "custom";
+    input.width = size.width;
+    input.height = size.height;
+  } else {
+    input.size = size;
+  }
+
+  if (referenceImages.length > 0) {
+    input.image_input = referenceImages;
+  }
+
+  if (args.aspectRatio) {
+    validateDocumentedAspectRatio(model, args.aspectRatio);
+    input.aspect_ratio = args.aspectRatio;
+  } else if (referenceImages.length > 0 && family === "seedream45") {
+    input.aspect_ratio = "match_input_image";
+  }
+
+  return input;
+}
+
+function buildWanInput(
+  family: "wan27image" | "wan27imagepro",
+  prompt: string,
+  args: CliArgs,
+  referenceImages: string[],
+): Record<string, unknown> {
+  const input: Record<string, unknown> = {
+    prompt,
+    size: resolveWanSize(family, args),
+  };
+
+  if (referenceImages.length > 0) {
+    input.images = referenceImages;
+  }
+
+  return input;
+}
+
+export function validateArgs(model: string, args: CliArgs): void {
+  parseModelId(model);
+
+  if (args.n !== 1) {
+    throw new Error("Replicate integration currently supports exactly one output image per request. Remove --n or use --n 1.");
+  }
+
+  if (args.imageSize && args.imageSizeSource !== "config") {
+    throw new Error("Replicate models in baoyu-image-gen do not use --imageSize. Use --quality, --ar, or --size instead.");
+  }
+
+  const family = getModelFamily(model);
+
+  if (family === "nano-banana") {
+    if (args.referenceImages.length > 14) {
+      throw new Error("Replicate nano-banana supports at most 14 reference images.");
+    }
+    if (args.aspectRatio) {
+      validateDocumentedAspectRatio(model, args.aspectRatio);
+    }
+    if (args.size) {
+      getRequestedAspectRatio(model, args);
+      getNanoBananaResolution(args);
+    }
+    return;
+  }
+
+  if (family === "seedream45") {
+    if (args.referenceImages.length > 14) {
+      throw new Error("Replicate Seedream 4.5 supports at most 14 reference images.");
+    }
+    if (args.aspectRatio) {
+      validateDocumentedAspectRatio(model, args.aspectRatio);
+    }
+    resolveSeedream45Size(args);
+    return;
+  }
+
+  if (family === "seedream5lite") {
+    if (args.referenceImages.length > 14) {
+      throw new Error("Replicate Seedream 5 Lite supports at most 14 reference images.");
+    }
+    if (args.aspectRatio) {
+      validateDocumentedAspectRatio(model, args.aspectRatio);
+    }
+    resolveSeedream5LiteSize(args);
+    return;
+  }
+
+  if (family === "wan27image" || family === "wan27imagepro") {
+    if (args.referenceImages.length > 9) {
+      throw new Error("Replicate Wan 2.7 image models support at most 9 reference images.");
+    }
+    if (args.aspectRatio) {
+      const parsed = parseAspectRatio(args.aspectRatio);
+      if (!parsed) {
+        throw new Error(`Replicate Wan aspect ratio must be in W:H format, got ${args.aspectRatio}.`);
+      }
+    }
+    resolveWanSize(family, args);
+    return;
+  }
+
+  const hasExplicitAspectRatio = !!args.aspectRatio && args.aspectRatioSource !== "config";
+
+  if (args.referenceImages.length > 0 || hasExplicitAspectRatio || args.size) {
+    throw new Error(
+      `Replicate model ${model} is not in the baoyu-image-gen compatibility list. Supported families: google/nano-banana*, bytedance/seedream-4.5, bytedance/seedream-5-lite, wan-video/wan-2.7-image, wan-video/wan-2.7-image-pro.`
+    );
+  }
+}
+
+export function getDefaultOutputExtension(model: string): ".png" {
+  const _family = getModelFamily(model);
+  return ".png";
+}
+
+export function buildInput(
+  model: string,
+  prompt: string,
+  args: CliArgs,
+  referenceImages: string[],
+): Record<string, unknown> {
+  const family = getModelFamily(model);
+
+  if (family === "nano-banana") {
+    return buildNanoBananaInput(prompt, model, args, referenceImages);
+  }
+
+  if (family === "seedream45" || family === "seedream5lite") {
+    return buildSeedreamInput(family, prompt, model, args, referenceImages);
+  }
+
+  if (family === "wan27image" || family === "wan27imagepro") {
+    return buildWanInput(family, prompt, args, referenceImages);
+  }
+
+  return { prompt };
+}
+
 async function readImageAsDataUrl(p: string): Promise<string> {
  const buf = await readFile(p);
  const ext = path.extname(p).toLowerCase();
@@ -150,6 +555,11 @@ export function extractOutputUrl(prediction: PredictionResponse): string {
  if (typeof output === "string") return output;

  if (Array.isArray(output)) {
+    if (output.length !== 1) {
+      throw new Error(
+        `Replicate returned ${output.length} outputs, but baoyu-image-gen currently supports saving exactly one image per request.`
+      );
+    }
    const first = output[0];
    if (typeof first === "string") return first;
  }
@@ -178,13 +588,14 @@ export async function generateImage(
  if (!apiToken) throw new Error("REPLICATE_API_TOKEN is required. Get one at https://replicate.com/account/api-tokens");

  const parsedModel = parseModelId(model);
+  validateArgs(model, args);

  const refDataUrls: string[] = [];
  for (const refPath of args.referenceImages) {
    refDataUrls.push(await readImageAsDataUrl(refPath));
  }

-  const input = buildInput(prompt, args, refDataUrls);
+  const input = buildInput(model, prompt, args, refDataUrls);

  console.log(`Generating image with Replicate (${model})...`);

@@ -25,6 +25,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -25,6 +25,7 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
    size: null,
    quality: null,
    imageSize: null,
+    imageApiDialect: null,
    referenceImages: [],
    n: 1,
    batchFile: null,
@@ -3,13 +3,14 @@ export type Provider =
  | "openai"
  | "openrouter"
  | "dashscope"
+  | "zai"
  | "minimax"
  | "replicate"
  | "jimeng"
  | "seedream"
-  | "azure"
-  | "zai";
+  | "azure";
 export type Quality = "normal" | "2k";
+export type OpenAIImageApiDialect = "openai-native" | "ratio-metadata";

 export type CliArgs = {
  prompt: string | null;
@@ -18,9 +19,12 @@ export type CliArgs = {
  provider: Provider | null;
  model: string | null;
  aspectRatio: string | null;
+  aspectRatioSource?: "cli" | "task" | "config" | null;
  size: string | null;
  quality: Quality | null;
  imageSize: string | null;
+  imageSizeSource?: "cli" | "task" | "config" | null;
+  imageApiDialect: OpenAIImageApiDialect | null;
  referenceImages: string[];
  n: number;
  batchFile: string | null;
@@ -40,6 +44,7 @@ export type BatchTaskInput = {
  size?: string | null;
  quality?: Quality | null;
  imageSize?: "1K" | "2K" | "4K" | null;
+  imageApiDialect?: OpenAIImageApiDialect | null;
  ref?: string[];
  n?: number;
 };
@@ -57,17 +62,18 @@ export type ExtendConfig = {
  default_quality: Quality | null;
  default_aspect_ratio: string | null;
  default_image_size: "1K" | "2K" | "4K" | null;
+  default_image_api_dialect: OpenAIImageApiDialect | null;
  default_model: {
    google: string | null;
    openai: string | null;
    openrouter: string | null;
    dashscope: string | null;
+    zai: string | null;
    minimax: string | null;
    replicate: string | null;
    jimeng: string | null;
    seedream: string | null;
    azure: string | null;
-    zai: string | null;
  };
  batch?: {
    max_workers?: number | null;