seahi/baoyu-skills
1
0
Fork 0
mirror of https://github.com/JimLiu/baoyu-skills.git synced 2026-08-08 09:53:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improve baoyu-imagine Replicate compatibility (#125)

Browse Source 
* Align Replicate image behavior with the models we actually support

Replicate image generation in baoyu-imagine no longer assumes that every model
accepts the nano-banana request schema. The Replicate provider now defaults to
google/nano-banana-2, routes supported model families through family-specific
builders and validators, blocks misleading multi-output requests before they
reach the API, and updates user-facing docs/config guidance to match the actual
contract.

Constraint: Replicate model families expose different input schemas
Constraint: Current Replicate path only saves one output image per request
Constraint: Must not change non-Replicate providers
Rejected: Keep one nano-banana-style payload for all Replicate models | triggers remote schema errors on Seedream and Wan
Rejected: Continue accepting multi-image Replicate requests and save only the first result | silently drops outputs
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Add a family-specific validator and input builder before exposing more Replicate model IDs or multi-output flags
Tested: npm test
Tested: node --test skills/baoyu-imagine/scripts/providers/replicate.test.ts skills/baoyu-imagine/scripts/main.test.ts
Not-tested: Live Replicate API calls against production models
Co-authored-by: justnode <justnode@users.noreply.github.com>

* Preserve Replicate compatibility when shared defaults leak across providers

Addressed the new PR review findings by teaching baoyu-imagine to track
where aspect-ratio defaults came from, mirroring the earlier imageSize fix,
so unsupported Replicate models can still run prompt-only requests when the
value was inherited from shared config. Also corrected Seedream 4.5 custom
size encoding to use the API's custom width/height schema instead of sending
literal WxH strings.

Constraint: Shared EXTEND defaults still need to apply globally for providers that support them
Constraint: Seedream 4.5 custom sizes must follow Replicate's documented custom size schema
Rejected: Ignore all aspect ratios for unknown Replicate models | would hide explicit unsupported CLI/task input
Rejected: Keep Seedream custom sizes as literal strings | validated locally but fails against the provider API
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Any future inherited-default validation for provider-specific flags should record the source explicitly before rejecting it
Tested: node --import tsx --test skills/baoyu-imagine/scripts/main.test.ts skills/baoyu-imagine/scripts/providers/replicate.test.ts
Tested: npm test
Not-tested: Live Replicate API calls for Seedream 4.5 custom-size requests

---------

Co-authored-by: justnode <justnode@users.noreply.github.com>
This commit is contained in:
Jim Liu 宝玉
2026-04-12 01:16:32 -05:00
committed by GitHub
parent 0b8ac256f4
commit 67a45a57a0
10 changed files with 772 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
skills/baoyu-imagine/SKILL.md
+38 -12
View File
@@ -76,7 +76,7 @@ ${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --quality 2k
# From prompt files
${BUN_X} {baseDir}/scripts/main.ts --promptfiles system.md content.md --image out.png
# With reference images (Google, OpenAI, Azure OpenAI, OpenRouter, Replicate, MiniMax, or Seedream 4.0/4.5/5.0)
# With reference images (Google, OpenAI, Azure OpenAI, OpenRouter, Replicate supported families, MiniMax, or Seedream 4.0/4.5/5.0)
${BUN_X} {baseDir}/scripts/main.ts --prompt "Make blue" --image out.png --ref source.png
# With reference images (explicit provider/model)
@@ -118,11 +118,14 @@ ${BUN_X} {baseDir}/scripts/main.ts --prompt "A girl stands by the library window
# MiniMax with custom size (documented for image-01)
${BUN_X} {baseDir}/scripts/main.ts --prompt "A cinematic poster" --image out.jpg --provider minimax --model image-01 --size 1536x1024
# Replicate (google/nano-banana-pro)
# Replicate (default: google/nano-banana-2)
${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider replicate
# Replicate with specific model
${BUN_X} {baseDir}/scripts/main.ts --prompt "A cat" --image out.png --provider replicate --model google/nano-banana
# Replicate Seedream 4.5
${BUN_X} {baseDir}/scripts/main.ts --prompt "A cinematic portrait" --image out.png --provider replicate --model bytedance/seedream-4.5 --ar 3:2
# Replicate Wan 2.7 Image Pro
${BUN_X} {baseDir}/scripts/main.ts --prompt "A concept frame" --image out.png --provider replicate --model wan-video/wan-2.7-image-pro --size 2048x1152
# Batch mode with saved prompt files
${BUN_X} {baseDir}/scripts/main.ts --batchfile batch.json
@@ -142,7 +145,7 @@ ${BUN_X} {baseDir}/scripts/main.ts --batchfile batch.json --jobs 4 --json
"promptFiles": ["prompts/hero.md"],
"image": "out/hero.png",
"provider": "replicate",
"model": "google/nano-banana-pro",
"model": "google/nano-banana-2",
"ar": "16:9",
"quality": "2k"
},
@@ -173,8 +176,8 @@ Paths in `promptFiles`, `image`, and `ref` are resolved relative to the batch fi
| `--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. Supported by Google multimodal, OpenAI GPT Image edits, Azure OpenAI edits (PNG/JPG only), OpenRouter multimodal models, Replicate, MiniMax subject-reference, and Seedream 5.0/4.5/4.0. Not supported by Jimeng, Seedream 3.0, or removed SeedEdit 3.0 |
| `--n <count>` | Number of images |
| `--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, and Seedream 5.0/4.5/4.0. Not supported by Jimeng, Seedream 3.0, or removed SeedEdit 3.0 |
| `--n <count>` | Number of images. Replicate currently supports only `--n 1` because this path saves exactly one output image |
| `--json` | JSON output |
## Environment Variables
@@ -202,7 +205,7 @@ Paths in `promptFiles`, `image`, and `ref` are resolved relative to the batch fi
| `ZAI_IMAGE_MODEL` | Z.AI model override (default: `glm-image`) |
| `BIGMODEL_IMAGE_MODEL` | Backward-compatible alias for Z.AI model override |
| `MINIMAX_IMAGE_MODEL` | MiniMax model override (default: `image-01`) |
| `REPLICATE_IMAGE_MODEL` | Replicate model override (default: google/nano-banana-pro) |
| `REPLICATE_IMAGE_MODEL` | Replicate model override (default: google/nano-banana-2) |
| `JIMENG_IMAGE_MODEL` | Jimeng model override (default: jimeng_t2i_v40) |
| `SEEDREAM_IMAGE_MODEL` | Seedream model override (default: doubao-seedream-5-0-260128) |
| `OPENAI_BASE_URL` | Custom OpenAI endpoint |
@@ -360,10 +363,33 @@ Notes:
### Replicate Models
Supported model formats:
Replicate support in `baoyu-imagine` is intentionally scoped to the model families that the tool can validate locally and save without dropping outputs:
- `owner/name` (recommended for official models), e.g. `google/nano-banana-pro`
- `owner/name:version` (community models by version), e.g. `stability-ai/sdxl:<version>`
- `google/nano-banana*` (default: `google/nano-banana-2`)
- Supports prompt-only and reference-image generation
- Uses Replicate `aspect_ratio`, `resolution`, and `output_format`
- `--size <WxH>` is accepted only as a shorthand for a documented aspect ratio plus `1K` / `2K`
- `bytedance/seedream-4.5`
- Supports prompt-only and reference-image generation
- Uses Replicate `size`, `aspect_ratio`, and `image_input`
- Local validation blocks unsupported `1K` requests before the API call
- `bytedance/seedream-5-lite`
- Supports prompt-only and reference-image generation
- Uses Replicate `size`, `aspect_ratio`, and `image_input`
- Local validation currently accepts `2K` / `3K` only
- `wan-video/wan-2.7-image`
- Supports prompt-only and reference-image generation
- Uses Replicate `size` and `images`
- Max output size is 2K
- `wan-video/wan-2.7-image-pro`
- Supports prompt-only and reference-image generation
- Uses Replicate `size` and `images`
- 4K is allowed only for text-to-image; local validation blocks `4K + --ref`
Guardrails:
- Replicate currently supports only single-output save semantics in this tool. Keep `--n 1`.
- If a Replicate model is outside the compatibility list above, `baoyu-imagine` only treats it as prompt-only and rejects advanced local options instead of guessing a nano-banana-style schema.
Examples:
@@ -398,7 +424,7 @@ Supported: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `2.35:1`
- Google multimodal: uses `imageConfig.aspectRatio`
- OpenAI: maps to closest supported size
- OpenRouter: sends `imageGenerationOptions.aspect_ratio`; if only `--size <WxH>` is given, aspect ratio is inferred automatically
- Replicate: passes `aspect_ratio` to model; when `--ref` is provided without `--ar`, defaults to `match_input_image`
- Replicate: behavior is model-family-specific. `google/nano-banana*` uses `aspect_ratio`; `bytedance/seedream-*` uses documented Replicate aspect ratios; Wan 2.7 maps `--ar` to a concrete `size`
- MiniMax: sends official `aspect_ratio` values directly; if `--size <WxH>` is given without `--ar`, `width` / `height` are sent for `image-01`
## Generation Mode