feat(baoyu-imagine): add DashScope Wan 2.7 image model support (#141)

* feat(baoyu-imagine): add DashScope Wan 2.7 image model support Closes #139. Adds the new `wan2.7-image-pro` and `wan2.7-image` model family to the DashScope provider so users can call Wan 2.7 directly through the official Aliyun (Bailian) API instead of going through Replicate. - Register `wan2.7-image-pro` and `wan2.7-image` as a new `wan27` family in the DashScope provider with their own size resolution rules: pixel range `[768*768, 4096*4096]` for `wan2.7-image-pro` text-to-image, `[768*768, 2048*2048]` for `wan2.7-image-pro` with refs and for the base `wan2.7-image` model in any mode, with aspect ratios validated against the documented `[1:8, 8:1]` band. - Allow up to 9 reference images per request (image editing / multi-image fusion). Local files are inlined as base64 data URLs; `http(s)://` paths are forwarded as-is. Other DashScope models still reject `--ref` with a hint to switch to a wan2.7 model or another provider. - Drop `prompt_extend` from the request body for the Wan 2.7 family (not part of the Wan 2.7 API surface) and skip the Qwen-only negative prompt for this family. - Allow `--provider dashscope --ref ...` in `detectProvider` so users can opt into Wan 2.7 reference workflows, while keeping Wan 2.7 out of the auto-detect ref priority list. - Add provider, reference, and usage-example documentation, plus unit tests covering family routing, size derivation across the three pixel-budget modes, ratio rejection, explicit-size validation, and the new `--provider dashscope` ref opt-in path. Made-with: Cursor * fix(baoyu-imagine): force n=1 for DashScope wan2.7 to avoid silent multi-image billing Cross-checked the implementation against the official Wan 2.7 image generation & editing API reference and found that the API defaults `parameters.n` to 4 in non-collage mode (1-4 range, billed per image). baoyu-imagine has single-image save semantics — only the first image in the response is kept — so without an explicit `n: 1` users would silently pay for 3 discarded images per request. - Always send `parameters.n: 1` in the wan2.7 request body - Reject `--n > 1` for wan2.7 with a clear error pointing at the single-image save semantics - Add tests asserting the request body shape (n=1, no prompt_extend, no negative_prompt) and the --n>1 rejection - Document the defaults-vs-skill mismatch in the dashscope reference Made-with: Cursor * Fix DashScope Wan 2.7 review feedback
2026-07-28 12:49:48 +08:00 · 2026-04-25 14:54:08 -05:00
parent 31d728b505
commit 6d063734ae
8 changed files with 551 additions and 19 deletions
@@ -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-imagine. 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-imagine 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,
        },
      ],
    },