vidspotai-shared 1.0.124 → 1.0.126

package/lib/services/aiGen/helpers.js

@@ -3,6 +3,8 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.modelSupportsLastFrame = modelSupportsLastFrame;
 exports.getModelsSupportingLastFrame = getModelsSupportingLastFrame;
 exports.modelDeclaresField = modelDeclaresField;
+exports.getInputVideoBilling = getInputVideoBilling;
+exports.isModelRetired = isModelRetired;
 exports.validateParams = validateParams;
 const aiModels_1 = require("../../globals/aiModels");
 const errors_1 = require("../../utils/errors");
@@ -37,11 +39,50 @@ function modelDeclaresField(modelKey, field) {
     const cfg = aiModels_1.aiModelConfigs[modelKey];
     return !!cfg?.fields && field in cfg.fields;
 }
+/**
+ * Input-video billing for getCreditUsed. Providers that bill input seconds
+ * (Wan 3.0, HappyHorse edit, MiniMax, Seedance 2.0) charge for EVERY video the
+ * request carries — the source clip and each reference clip — so both count.
+ * An unmeasured clip falls back to the output duration (the same conservative
+ * proxy the source clip always used). Every billing call site (precheck,
+ * upfront deduction, multi-scene true-up, reconcile) must go through this so
+ * the estimate and the reconcile agree.
+ */
+function getInputVideoBilling(src) {
+    const numInputImages = (src.inputImageUrl ? 1 : 0) +
+        (src.lastFrameImageUrl ? 1 : 0) +
+        (src.referenceImageUrls?.length ?? 0);
+    const numInputAudios = src.referenceAudioUrls?.length ?? 0;
+    const fallback = src.duration ?? 0;
+    let seconds = 0;
+    let clips = 0;
+    if (src.inputVideoUrl) {
+        seconds += src.inputVideoDurationSec ?? fallback;
+        clips += 1;
+    }
+    (src.referenceVideoUrls ?? []).forEach((_, i) => {
+        seconds += src.referenceVideoDurationsSec?.[i] ?? fallback;
+        clips += 1;
+    });
+    // return clips ? { hasInputVideo: true, inputVideoDurationSec: seconds } : { hasInputVideo: false }; // REMOVED FOR v2 — now also counts input images
+    return clips
+        ? { hasInputVideo: true, inputVideoDurationSec: seconds, numInputImages, numInputAudios }
+        : { hasInputVideo: false, numInputImages, numInputAudios };
+}
+function isModelRetired(modelKey) {
+    const cfg = aiModels_1.aiModelConfigs[modelKey];
+    return !!cfg?.retired;
+}
 function validateParams(params) {
     const modelConfig = aiModels_1.aiModelConfigs[params.modelKey];
     if (!modelConfig) {
         throw new Error(`Unsupported model: ${params.modelKey}`);
     }
+    // 0. Retired upstream — fail fast with a user-facing, non-paging error
+    // instead of a billed round-trip to a dead endpoint.
+    if (modelConfig.retired) {
+        throw new errors_1.UserFacingError(`Model ${params.modelKey} is no longer available: ${modelConfig.retired.reason} Please pick a different model.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+    }
     const errors = [];
     // 1. Field-level validation (required + allowedValues)
     for (const [field, rules] of Object.entries(modelConfig.fields)) {
@@ -84,8 +125,13 @@ function validateParams(params) {
         (!modelConfig.type.includes("image-to-video") || !params.inputImageUrl)) {
         errors.push(`Model ${params.modelKey} requires imageUrl for image-to-video mode`);
     }
-    if (params.type === "video-to-video" &&
-        (modelConfig.type.includes("video-to-video") || !params.inputVideoUrl)) {
+    // REMOVED FOR v2 — kept for reference: inverted — it rejected every
+    // video-to-video request on a model that DOES declare video-to-video.
+    // if (
+    //   params.type === "video-to-video" &&
+    //   (modelConfig.type.includes("video-to-video") || !params.inputVideoUrl)
+    // ) {
+    if (params.type === "video-to-video" && !params.inputVideoUrl) {
         errors.push(`Model ${params.modelKey} requires videoUrl for video-to-video mode`);
     }
     if (modelConfig.type.includes("text-to-video") && !params.prompt) {

package/lib/services/aiGen/index.d.ts

@@ -3,4 +3,5 @@ export * from "./types";
 export * from "./helpers";
 export * from "./providers";
 export * from "./transientRetry";
+export { applyCameraDirection } from "./canonicalAdapters/cameraDirection";
 //# sourceMappingURL=index.d.ts.map

package/lib/services/aiGen/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/services/aiGen/index.ts"],"names":[],"mappings":"AAAA,cAAc,wBAAwB,CAAC;AACvC,cAAc,SAAS,CAAC;AACxB,cAAc,WAAW,CAAC;AAC1B,cAAc,aAAa,CAAC;AAC5B,cAAc,kBAAkB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/services/aiGen/index.ts"],"names":[],"mappings":"AAAA,cAAc,wBAAwB,CAAC;AACvC,cAAc,SAAS,CAAC;AACxB,cAAc,WAAW,CAAC;AAC1B,cAAc,aAAa,CAAC;AAC5B,cAAc,kBAAkB,CAAC;AACjC,OAAO,EAAE,oBAAoB,EAAE,MAAM,qCAAqC,CAAC"}

package/lib/services/aiGen/index.js

@@ -14,8 +14,11 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.applyCameraDirection = void 0;
 __exportStar(require("./aiGenFactory.service"), exports);
 __exportStar(require("./types"), exports);
 __exportStar(require("./helpers"), exports);
 __exportStar(require("./providers"), exports);
 __exportStar(require("./transientRetry"), exports);
+var cameraDirection_1 = require("./canonicalAdapters/cameraDirection");
+Object.defineProperty(exports, "applyCameraDirection", { enumerable: true, get: function () { return cameraDirection_1.applyCameraDirection; } });

package/lib/services/aiGen/providers/alibaba/alibaba.d.ts

@@ -1,5 +1,16 @@
 import { BaseAiGenProviderService } from "../baseAiGenProvider.service";
 import { CreditUsageParams, ImageGenerationParams, ImageGenerationResult, VideoGenerationParams, VideoGenerationResult, VideoStatusParams, VideoStatusResult } from "../types";
+/**
+ * Build the `{input, parameters}` body for the media-array models. Mode is
+ * driven by the model id (HappyHorse ships one model per mode) and, for the
+ * all-in-one Wan 3.0, by which inputs are present. Reference order matters:
+ * the prompt cites media as "Image 1" / "Video 1" / "Audio 1" (Wan 3.0) or
+ * "[Image 1]" (HappyHorse) in array order, per type.
+ */
+export declare function buildMediaArrayVideoBody(modelId: string, params: VideoGenerationParams): {
+    input: Record<string, unknown>;
+    parameters: Record<string, unknown>;
+};
 /**
  * Map a canonical "W:H" aspect ratio to a DashScope pixel `size` string.
  * Returns undefined for unparseable input so the caller simply omits `size`
@@ -17,6 +28,7 @@ export declare class AlibabaService extends BaseAiGenProviderService {
      */
     private request;
     generateVideo(params: VideoGenerationParams): Promise<VideoGenerationResult>;
+    private submitVideo;
     checkVideoStatus({ task, outputFilename, outputFilePath, modelKey, }: VideoStatusParams): Promise<VideoStatusResult>;
     /**
      * DashScope image generation. Three dispatch paths (verified live 2026-06-06):
@@ -51,6 +63,6 @@ export declare class AlibabaService extends BaseAiGenProviderService {
      * and text2image/image-synthesis (legacy Qwen).
      */
     private generateImageAsync;
-    getCreditUsed({ modelKey, resolution, aspectRatio, duration, multiClip, numImages, creditFloor, }: CreditUsageParams): number;
+    getCreditUsed({ modelKey, resolution, aspectRatio, duration, multiClip, numImages, creditFloor, hasInputVideo, inputVideoDurationSec, audio, numInputImages, }: CreditUsageParams): number;
 }
 //# sourceMappingURL=alibaba.d.ts.map

package/lib/services/aiGen/providers/alibaba/alibaba.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"alibaba.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/alibaba/alibaba.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAoGlB;;;;GAIG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,GAClB,MAAM,GAAG,SAAS,CAiBpB;AAED,qBAAa,cAAe,SAAQ,wBAAwB;IAE1D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAwB;IAChD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAS;;IAiBjC;;;;OAIG;YACW,OAAO;IAwDf,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA+E3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,EACzB,QAAQ,GACT,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAqEjD;;;;;;;;;;;;;;;;;;;;;OAqBG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA0EjC;;;OAGG;YACW,iBAAiB;IA4C/B;;;OAGG;YACW,kBAAkB;IA6DhC,aAAa,CAAC,EACZ,QAAQ,EACR,UAAU,EACV,WAAW,EACX,QAAQ,EACR,SAAiB,EACjB,SAAa,EACb,WAAW,GACZ,EAAE,iBAAiB,GAAG,MAAM;CAoB9B"}
+{"version":3,"file":"alibaba.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/alibaba/alibaba.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAwFlB;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACtC,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,qBAAqB,GAC5B;IAAE,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAgJzE;AA2CD;;;;GAIG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,GAClB,MAAM,GAAG,SAAS,CAmBpB;AAED,qBAAa,cAAe,SAAQ,wBAAwB;IAE1D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAwB;IAChD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAS;;IAiBjC;;;;OAIG;YACW,OAAO;IAwDf,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;YAyInB,WAAW;IAwBnB,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,EACzB,QAAQ,GACT,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAqEjD;;;;;;;;;;;;;;;;;;;;;OAqBG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA6EjC;;;OAGG;YACW,iBAAiB;IAoD/B;;;OAGG;YACW,kBAAkB;IA6DhC,aAAa,CAAC,EACZ,QAAQ,EACR,UAAU,EACV,WAAW,EACX,QAAQ,EACR,SAAiB,EACjB,SAAa,EACb,WAAW,EACX,aAAa,EACb,qBAAqB,EACrB,KAAK,EACL,cAAc,GACf,EAAE,iBAAiB,GAAG,MAAM;CA4C9B"}

package/lib/services/aiGen/providers/alibaba/alibaba.js

@@ -4,6 +4,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.AlibabaService = void 0;
+exports.buildMediaArrayVideoBody = buildMediaArrayVideoBody;
 exports.resolveDashScopeSize = resolveDashScopeSize;
 const axios_1 = __importDefault(require("axios"));
 const types_1 = require("../../../../globals/types");
@@ -29,7 +30,14 @@ const errors_1 = require("../../../../utils/errors");
 // S2V (wan2.2-s2v) is China-region only and cannot be reached with an intl
 // DashScope key; the model key is intentionally kept but rejected at submit
 // with a UserFacingError. Same for kolors-v2 (Kuaishou model, not Alibaba).
-const ALIBABA_BASE_URL = "https://dashscope-intl.aliyuncs.com/api/v1";
+//
+// Host: Alibaba now recommends the workspace domain
+// https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1 and says the
+// DashScope domain gets no NEW features after 2026-09-30 (existing models keep
+// working). ALIBABA_API_BASE_URL switches hosts without a code change; the
+// same API key works on both.
+// const ALIBABA_BASE_URL = "https://dashscope-intl.aliyuncs.com/api/v1"; // REMOVED FOR v2 — now env-overridable
+const ALIBABA_BASE_URL = (process.env.ALIBABA_API_BASE_URL || "https://dashscope-intl.aliyuncs.com/api/v1").replace(/\/+$/, "");
 const ALIBABA_BASE = `${ALIBABA_BASE_URL}/services/aigc`;
 const ENDPOINT_VIDEO_ASYNC = `${ALIBABA_BASE}/video-generation/video-synthesis`;
 const ENDPOINT_IMAGE_T2I_ASYNC = `${ALIBABA_BASE}/text2image/image-synthesis`;
@@ -53,6 +61,7 @@ const INTL_UNAVAILABLE_MODEL_IDS = new Set([
 //   - Legacy qwen-image, qwen-image-plus.
 function isSyncMultimodalImageModel(modelId) {
     return (modelId.startsWith("qwen-image-2") ||
+        modelId.startsWith("qwen-image-3") ||
         modelId.startsWith("qwen-image-max") ||
         modelId.startsWith("qwen-image-edit") ||
         modelId.startsWith("wan2.7-image"));
@@ -60,6 +69,165 @@ function isSyncMultimodalImageModel(modelId) {
 function isAsyncWanImageModel(modelId) {
     return modelId.startsWith("wan2.7-image") || modelId.startsWith("wan2.6-image");
 }
+/**
+ * Wan 3.0, HappyHorse and Wan 2.7 r2v / videoedit take every input as
+ * `input.media[{type,url}]` and pick size with `resolution` ("1080P") +
+ * `ratio` instead of a pixel `size`. Only the Wan 2.7 pair take a negative prompt. HappyHorse also watermarks by DEFAULT, so we always
+ * send an explicit `watermark`.
+ */
+function isMediaArrayVideoModel(modelId) {
+    // return modelId.startsWith("wan3") || modelId.startsWith("happyhorse"); // REMOVED FOR v2 — + Wan 2.7 r2v / videoedit
+    return (modelId.startsWith("wan3") ||
+        modelId.startsWith("happyhorse") ||
+        modelId.startsWith("wan2.7-r2v") ||
+        modelId.startsWith("wan2.7-videoedit"));
+}
+/**
+ * Build the `{input, parameters}` body for the media-array models. Mode is
+ * driven by the model id (HappyHorse ships one model per mode) and, for the
+ * all-in-one Wan 3.0, by which inputs are present. Reference order matters:
+ * the prompt cites media as "Image 1" / "Video 1" / "Audio 1" (Wan 3.0) or
+ * "[Image 1]" (HappyHorse) in array order, per type.
+ */
+function buildMediaArrayVideoBody(modelId, params) {
+    const media = [];
+    const refImages = params.referenceImageUrls ?? [];
+    const refVideos = [
+        ...(params.inputVideoUrl ? [params.inputVideoUrl] : []),
+        ...(params.referenceVideoUrls ?? []),
+    ];
+    const refAudios = params.referenceAudioUrls ?? [];
+    const isHappyHorse = modelId.startsWith("happyhorse");
+    const isVideoEdit = modelId === "happyhorse-1.0-video-edit";
+    // const isHappyHorseI2V = modelId === "happyhorse-1.1-i2v"; // REMOVED FOR v2 — suffix covers 1.0 + 1.1
+    // HappyHorse ships one model per mode; the id suffix names it (-t2v/-i2v/-r2v).
+    const isHappyHorseI2V = isHappyHorse && modelId.endsWith("-i2v");
+    const isWan27R2V = modelId.startsWith("wan2.7-r2v");
+    const isWan27VideoEdit = modelId.startsWith("wan2.7-videoedit");
+    const hasFrames = !!params.inputImageUrl || !!params.lastFrameImageUrl;
+    const hasDocOrLink = !!params.referenceDocumentUrl || !!params.referenceWebPageUrl;
+    const hasRefs = refImages.length + refVideos.length + refAudios.length > 0 || hasDocOrLink;
+    if (isWan27R2V) {
+        // ≤1 first frame + 1–5 subjects (images + videos combined); each subject
+        // may carry a voice. Voices map in send order: videos first, then images.
+        const subjects = [
+            ...(params.referenceVideoUrls ?? []).map((url) => ({ type: "reference_video", url })),
+            ...refImages.map((url) => ({ type: "reference_image", url })),
+        ];
+        if (subjects.length < 1 || subjects.length > 5) {
+            throw new errors_1.UserFacingError(`${modelId} needs 1–5 reference images and videos in total (got ${subjects.length}).`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        if (refAudios.length > subjects.length) {
+            throw new errors_1.UserFacingError(`${modelId} takes one voice per reference subject: ${refAudios.length} voices for ${subjects.length} subjects.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        if ((params.referenceVideoUrls?.length ?? 0) > 0 && (params.duration ?? 5) > 10) {
+            throw new errors_1.UserFacingError(`${modelId} is limited to 10s when a reference video is used.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        if (params.inputImageUrl)
+            media.push({ type: "first_frame", url: params.inputImageUrl });
+        subjects.forEach((m, i) => media.push(refAudios[i] ? { ...m, reference_voice: refAudios[i] } : m));
+    }
+    else if (isWan27VideoEdit) {
+        if (!params.inputVideoUrl) {
+            throw new errors_1.UserFacingError(`${modelId} needs the video to edit (inputVideoUrl).`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        media.push({ type: "video", url: params.inputVideoUrl });
+        for (const url of refImages.slice(0, 4))
+            media.push({ type: "reference_image", url });
+    }
+    else if (isVideoEdit) {
+        if (!params.inputVideoUrl) {
+            throw new errors_1.UserFacingError(`${modelId} needs the video to edit (inputVideoUrl).`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        media.push({ type: "video", url: params.inputVideoUrl });
+        for (const url of refImages.slice(0, 5))
+            media.push({ type: "reference_image", url });
+    }
+    else if (isHappyHorseI2V) {
+        media.push({ type: "first_frame", url: params.inputImageUrl });
+    }
+    else if (isHappyHorse) {
+        // r2v: 1–9 reference images; t2v: no media.
+        for (const url of refImages.slice(0, 9))
+            media.push({ type: "reference_image", url });
+    }
+    else {
+        // Wan 3.0: first/last frame XOR references (the API rejects mixing them).
+        if (hasFrames && hasRefs) {
+            throw new errors_1.UserFacingError(`${modelId} can't combine first/last frame images with reference images, videos, audio, documents or web pages. Use one or the other.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        if (params.referenceDocumentUrl && params.referenceWebPageUrl) {
+            throw new errors_1.UserFacingError(`${modelId} takes a reference document OR a web page, not both.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
+        if (params.inputImageUrl)
+            media.push({ type: "first_frame", url: params.inputImageUrl });
+        if (params.lastFrameImageUrl)
+            media.push({ type: "last_frame", url: params.lastFrameImageUrl });
+        for (const url of refImages.slice(0, 10))
+            media.push({ type: "reference_image", url });
+        for (const url of refVideos.slice(0, 5))
+            media.push({ type: "reference_video", url });
+        for (const url of refAudios.slice(0, 5))
+            media.push({ type: "reference_audio", url });
+        if (params.referenceDocumentUrl)
+            media.push({ type: "file", url: params.referenceDocumentUrl });
+        if (params.referenceWebPageUrl)
+            media.push({ type: "link", url: params.referenceWebPageUrl });
+    }
+    const input = {};
+    if (params.prompt)
+        input.prompt = params.prompt;
+    if (media.length)
+        input.media = media;
+    // Wan 2.7 takes a negative prompt; Wan 3.0 / HappyHorse don't.
+    if ((isWan27R2V || isWan27VideoEdit) && params.negativePrompt) {
+        input.negative_prompt = params.negativePrompt;
+    }
+    const parameters = {
+        watermark: params.watermark ?? false,
+    };
+    if (params.resolution)
+        parameters.resolution = params.resolution.toUpperCase();
+    if (params.seed !== undefined)
+        parameters.seed = params.seed;
+    if (isWan27VideoEdit) {
+        // Output = the whole input clip by default; `duration` truncates from 0s,
+        // so it's only sent when we know it's shorter than the clip.
+        if (params.duration &&
+            params.inputVideoDurationSec !== undefined &&
+            params.duration < params.inputVideoDurationSec) {
+            parameters.duration = params.duration;
+        }
+        if (params.aspectRatio)
+            parameters.ratio = params.aspectRatio;
+        parameters.audio_setting = params.audio === false ? "origin" : "auto";
+        parameters.prompt_extend = params.promptOptimizer ?? true;
+        return { input, parameters };
+    }
+    if (isVideoEdit) {
+        // Output length follows the input clip; no duration/ratio. `audio: false`
+        // keeps the source audio, otherwise the model decides.
+        parameters.audio_setting = params.audio === false ? "origin" : "auto";
+        return { input, parameters };
+    }
+    parameters.duration = params.duration || 5;
+    // Frame-driven output follows the image, so ratio is only sent without a
+    // first frame (Wan 3.0 defaults to "adaptive"; HappyHorse i2v has no ratio).
+    if (params.aspectRatio && !params.inputImageUrl && !isHappyHorseI2V) {
+        parameters.ratio = params.aspectRatio;
+    }
+    if (isWan27R2V) {
+        // Always generates audio; no `audio` switch.
+        parameters.prompt_extend = params.promptOptimizer ?? true;
+    }
+    else if (!isHappyHorse) {
+        if (params.audio !== undefined)
+            parameters.audio = params.audio;
+        // Document / web-page inputs require prompt_extend=true (API rule).
+        parameters.prompt_extend = hasDocOrLink ? true : params.promptOptimizer ?? true;
+    }
+    return { input, parameters };
+}
 /**
  * DashScope's `size` parameter is "WIDTH*HEIGHT" in PIXELS — it is NOT an
  * aspect-ratio string. We used to assign `parameters.size = params.aspectRatio`
@@ -105,7 +273,8 @@ function round16(n) {
  * and lets DashScope apply its own default (better than sending a bad value).
  */
 function resolveDashScopeSize(modelId, aspectRatio) {
-    const family = modelId.startsWith("qwen-image-2") ? "large" : "standard";
+    // REMOVED FOR v2 — kept for reference: const family = modelId.startsWith("qwen-image-2") ? "large" : "standard";
+    const family = modelId.startsWith("qwen-image-2") || modelId.startsWith("qwen-image-3") ? "large" : "standard";
     const preset = DASHSCOPE_SIZE_PRESETS[family][aspectRatio];
     if (preset)
         return preset;
@@ -192,52 +361,131 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
         if (INTL_UNAVAILABLE_MODEL_IDS.has(modelId)) {
             throw new errors_1.UserFacingError(`Model "${modelId}" is not available on the intl DashScope endpoint (dashscope-intl.aliyuncs.com). It is hosted only in the China region and requires a separate Beijing-region API key.`, errors_1.USER_FACING_ERROR_CODES.PROVIDER_AUTH_ERROR);
         }
-        // T2V and I2V share a single endpoint. The model + presence of img_url
-        // determines mode. S2V is unavailable on intl (gated above).
-        const isI2V = !!params.inputImageUrl;
+        if (isMediaArrayVideoModel(modelId)) {
+            const { input, parameters } = buildMediaArrayVideoBody(modelId, params);
+            return this.submitVideo({ model: modelId, input, parameters });
+        }
+        // REMOVED FOR v2 — kept for reference. Sent a pixel `size` to wan2.7-t2v
+        // (which replaced it with resolution + ratio), never sent `resolution` for
+        // I2V (so every I2V ran at the 1080P default whatever we billed), and
+        // ignored first_clip / driving audio.
+        // // T2V and I2V share a single endpoint. The model + presence of img_url
+        // // determines mode. S2V is unavailable on intl (gated above).
+        // const isI2V = !!params.inputImageUrl;
+        // const input: Record<string, unknown> = {};
+        // if (params.prompt) input.prompt = params.prompt;
+        // if (params.negativePrompt) input.negative_prompt = params.negativePrompt;
+        // if (isI2V) {
+        //   // wan2.7-i2v moved to a media[] array ({type,url}) instead of the
+        //   // legacy img_url/last_frame_url fields — confirmed against Alibaba's
+        //   // "Wan 2.7 - image-to-video API" reference (2026-08-20); wan2.6-i2v(-flash)
+        //   // and older versions still use img_url/last_frame_url.
+        //   if (modelId.startsWith("wan2.7")) {
+        //     const media: Array<{ type: string; url: string }> = [
+        //       { type: "first_frame", url: params.inputImageUrl! },
+        //     ];
+        //     if (params.lastFrameImageUrl) {
+        //       media.push({ type: "last_frame", url: params.lastFrameImageUrl });
+        //     }
+        //     input.media = media;
+        //   } else {
+        //     input.img_url = params.inputImageUrl;
+        //     // wan2.6-i2v supports first-last-frame interpolation
+        //     if (params.lastFrameImageUrl) input.last_frame_url = params.lastFrameImageUrl;
+        //   }
+        // }
+        // const parameters: Record<string, unknown> = {
+        //   duration: params.duration || 5,
+        //   prompt_extend: params.promptOptimizer ?? true,
+        // };
+        // // T2V requires explicit size; I2V derives dimensions from the input image.
+        // if (!isI2V) {
+        //   const size = getAlibabaDimensions(params.resolution!, params.aspectRatio!);
+        //   if (!size) {
+        //     throw new Error(
+        //       `Invalid resolution/aspect ratio combination: ${params.resolution} ${params.aspectRatio}`
+        //     );
+        //   }
+        //   parameters.size = size;
+        // }
+        // if (params.seed !== undefined) parameters.seed = params.seed;
+        // if (params.watermark !== undefined) parameters.watermark = params.watermark;
+        // T2V and I2V share a single endpoint. The model + inputs determine mode.
+        const isWan27 = modelId.startsWith("wan2.7");
+        const isI2V = !!params.inputImageUrl || (isWan27 && !!params.inputVideoUrl);
+        const resolutionTier = params.resolution ? params.resolution.toUpperCase() : undefined;
+        if (modelId.startsWith("wan2.7-i2v") && !isI2V) {
+            throw new errors_1.UserFacingError(`${modelId} needs a first-frame image or a first video clip to continue.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
         const input = {};
         if (params.prompt)
             input.prompt = params.prompt;
         if (params.negativePrompt)
             input.negative_prompt = params.negativePrompt;
-        if (isI2V) {
-            // wan2.7-i2v moved to a media[] array ({type,url}) instead of the
-            // legacy img_url/last_frame_url fields — confirmed against Alibaba's
-            // "Wan 2.7 - image-to-video API" reference (2026-08-20); wan2.6-i2v(-flash)
-            // and older versions still use img_url/last_frame_url.
-            if (modelId.startsWith("wan2.7")) {
-                const media = [
-                    { type: "first_frame", url: params.inputImageUrl },
-                ];
-                if (params.lastFrameImageUrl) {
-                    media.push({ type: "last_frame", url: params.lastFrameImageUrl });
-                }
-                input.media = media;
-            }
-            else {
-                input.img_url = params.inputImageUrl;
-                // wan2.6-i2v supports first-last-frame interpolation
-                if (params.lastFrameImageUrl)
-                    input.last_frame_url = params.lastFrameImageUrl;
-            }
-        }
         const parameters = {
             duration: params.duration || 5,
             prompt_extend: params.promptOptimizer ?? true,
         };
-        // T2V requires explicit size; I2V derives dimensions from the input image.
-        if (!isI2V) {
+        if (isI2V && isWan27) {
+            // wan2.7-i2v: everything in media[] ({type,url}); each type at most once.
+            //   first_frame [+ last_frame] [+ driving_audio]  or  first_clip [+ last_frame]
+            // Output aspect follows the first frame / clip, so no ratio.
+            const media = [];
+            if (params.inputVideoUrl) {
+                media.push({ type: "first_clip", url: params.inputVideoUrl });
+            }
+            else {
+                media.push({ type: "first_frame", url: params.inputImageUrl });
+            }
+            if (params.lastFrameImageUrl) {
+                media.push({ type: "last_frame", url: params.lastFrameImageUrl });
+            }
+            if (params.inputAudioUrl && !params.inputVideoUrl) {
+                media.push({ type: "driving_audio", url: params.inputAudioUrl });
+            }
+            input.media = media;
+            if (resolutionTier)
+                parameters.resolution = resolutionTier;
+        }
+        else if (isI2V) {
+            // wan2.6 and older: img_url / last_frame_url, `resolution` tier.
+            input.img_url = params.inputImageUrl;
+            // REMOVED FOR v2 — the legacy I2V API has no last_frame_url (only wan2.7 does FLF):
+            // if (params.lastFrameImageUrl) input.last_frame_url = params.lastFrameImageUrl;
+            if (params.inputAudioUrl)
+                input.audio_url = params.inputAudioUrl;
+            if (resolutionTier)
+                parameters.resolution = resolutionTier;
+        }
+        else if (isWan27) {
+            // wan2.7-t2v: resolution + ratio replace the pixel `size`.
+            if (params.inputAudioUrl)
+                input.audio_url = params.inputAudioUrl;
+            if (resolutionTier)
+                parameters.resolution = resolutionTier;
+            if (params.aspectRatio)
+                parameters.ratio = params.aspectRatio;
+        }
+        else {
+            // wan2.6-t2v and older T2V: pixel `size` ("1920*1080").
+            if (params.inputAudioUrl)
+                input.audio_url = params.inputAudioUrl;
             const size = (0, helpers_1.getAlibabaDimensions)(params.resolution, params.aspectRatio);
             if (!size) {
                 throw new Error(`Invalid resolution/aspect ratio combination: ${params.resolution} ${params.aspectRatio}`);
             }
             parameters.size = size;
         }
+        if (params.audio !== undefined && modelId === "wan2.6-i2v-flash") {
+            parameters.audio = params.audio; // silent video is half price
+        }
         if (params.seed !== undefined)
             parameters.seed = params.seed;
         if (params.watermark !== undefined)
             parameters.watermark = params.watermark;
-        const body = { model: modelId, input, parameters };
+        return this.submitVideo({ model: modelId, input, parameters });
+    }
+    async submitVideo(body) {
         const result = await this.request(body, "POST", ENDPOINT_VIDEO_ASYNC, true);
         // DashScope returns { output: { task_id, ... } } on async submit.
         const taskId = result?.output?.task_id || result?.request_id;
@@ -341,7 +589,10 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
             ...(params.inputImageUrls ?? []),
         ];
         const isEdit = modelId.startsWith("qwen-image-edit") ||
-            (refs.length > 0 && (modelId.startsWith("wan2.7-image") || modelId.startsWith("qwen-image-2")));
+            (refs.length > 0 &&
+                (modelId.startsWith("wan2.7-image") ||
+                    modelId.startsWith("qwen-image-2") ||
+                    modelId.startsWith("qwen-image-3")));
         if (isEdit && !refs.length) {
             throw new Error(`${modelId}: edit/reference mode requires at least one input image`);
         }
@@ -406,6 +657,12 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
             parameters.prompt_extend = params.promptOptimizer;
         if (params.seed !== undefined)
             parameters.seed = params.seed;
+        // Qwen-Image 2.x / 3.x return up to 6 images per call. `n` was never sent,
+        // so a multi-image request was billed for N and returned one.
+        if ((modelId.startsWith("qwen-image-2") || modelId.startsWith("qwen-image-3")) &&
+            (params.numImages ?? 1) > 1) {
+            parameters.n = Math.min(6, params.numImages);
+        }
         // `size` is WIDTH*HEIGHT pixels, never a ratio string — see
         // resolveDashScopeSize. An explicit imageSize (already in pixel form) wins.
         if (params.aspectRatio) {
@@ -475,11 +732,34 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
         const urls = results.map((r) => r.url).filter((u) => !!u);
         return { urls, taskId };
     }
-    getCreditUsed({ modelKey, resolution, aspectRatio, duration, multiClip = false, numImages = 1, creditFloor, }) {
+    getCreditUsed({ modelKey, resolution, aspectRatio, duration, multiClip = false, numImages = 1, creditFloor, hasInputVideo, inputVideoDurationSec, audio, numInputImages, }) {
         const modelConfig = aiModels_1.aiModelConfigs[modelKey];
-        // Image-gen — flat per-image cost (Kolors / Qwen-Image).
+        // Wan 3.0 / HappyHorse — per second by output resolution, plus input
+        // video seconds where the model bills them (Wan 3.0 references, video edit).
+        const perSecond = modelConfig?.cost?.perSecondByResolution;
+        if (perSecond) {
+            const res = resolution && resolution.toLowerCase() in perSecond
+                ? resolution.toLowerCase()
+                : String(modelConfig.fields.resolution?.default ?? "");
+            const silent = modelConfig.cost?.silentPerSecondByResolution;
+            const rate = (audio === false ? silent?.[res] : undefined) ?? perSecond[res];
+            if (rate !== undefined) {
+                const seconds = duration ?? Number(modelConfig.fields.duration?.default ?? 5);
+                const videoInRate = modelConfig.cost?.inputVideoPerSecond?.[res] ?? 0;
+                // Some models cap billable input seconds (Wan 2.7 r2v: 5s total).
+                const maxIn = typeof modelConfig.cost?.inputVideoBillableMaxSec === "number"
+                    ? modelConfig.cost.inputVideoBillableMaxSec
+                    : Infinity;
+                // const videoIn = hasInputVideo ? videoInRate * (inputVideoDurationSec ?? 0) : 0; // REMOVED FOR v2 — uncapped
+                const videoIn = hasInputVideo ? videoInRate * Math.min(inputVideoDurationSec ?? 0, maxIn) : 0;
+                return (0, utils_1.getCreditsFromCost)(rate * seconds + videoIn, !multiClip);
+            }
+        }
+        // Image-gen — flat per-image cost (Kolors / Qwen-Image), plus input
+        // images where the model bills them (Qwen-Image 3.0: $0.003 each).
         if (modelConfig?.cost?.fixed !== undefined) {
-            return (0, utils_1.getCreditsFromCost)(modelConfig.cost.fixed * Math.max(1, numImages), !multiClip, creditFloor);
+            const perInput = typeof modelConfig.cost.perInputImage === "number" ? modelConfig.cost.perInputImage : 0;
+            return (0, utils_1.getCreditsFromCost)(modelConfig.cost.fixed * Math.max(1, numImages) + perInput * (numInputImages ?? 0), !multiClip, creditFloor);
         }
         const dimensions = (0, helpers_1.getAlibabaDimensions)(resolution, aspectRatio);
         const cost = modelConfig.cost?.table?.[dimensions]?.[duration];

package/lib/services/aiGen/providers/bytedance/bytedance.service.d.ts

@@ -1,11 +1,39 @@
 import { BaseAiGenProviderService } from "../baseAiGenProvider.service";
-import { CreditUsageParams, VideoGenerationParams, VideoGenerationResult, VideoStatusParams, VideoStatusResult } from "../types";
+import { CreditUsageParams, ImageGenerationParams, ImageGenerationResult, VideoGenerationParams, VideoGenerationResult, VideoStatusParams, VideoStatusResult } from "../types";
+/**
+ * Explicit "WxH" for a ratio at a size tier: tier² pixels in the requested
+ * shape, floored to multiples of 16. Every tier's pixel count sits inside the
+ * range the models that offer it accept (e.g. 5.0 lite: [2560×1440, 4096²]),
+ * and flooring never drops below it, so any ratio in [1/16, 16] is valid —
+ * including ones a model's aspect list doesn't name (the agent's 4:5).
+ */
+export declare function seedreamSize(aspectRatio: string, tier: string): string | undefined;
 export declare class ByteDanceService extends BaseAiGenProviderService {
     private readonly baseUrl;
     private sanitizePrompt;
     private isSeedance2;
     generateVideo(params: VideoGenerationParams): Promise<VideoGenerationResult>;
     checkVideoStatus({ task, outputFilename, outputFilePath, }: VideoStatusParams): Promise<VideoStatusResult>;
+    /**
+     * Seedream text-to-image / image-to-image. Three modes, each gated on the
+     * model declaring the field:
+     *   - default: one request per output image, run in parallel, so numImages
+     *     is exact.
+     *   - `imageSet` (5.0 lite / 4.x): one `sequential_image_generation: "auto"`
+     *     request returning a set of related images, at most numImages (the model
+     *     picks the count; the worker refunds the unused share).
+     *   - `layerDecomposition` (5.0 pro / flash): one input image → base image +
+     *     up to 16 transparent PNG layers, returned with their stacking order and
+     *     bounding boxes.
+     */
+    generateImage(params: ImageGenerationParams): Promise<ImageGenerationResult>;
+    /** Copy a layer decomposition's outputs into our storage, base image first. */
+    private storeSeedreamLayers;
+    /**
+     * One /images/generations call → its successful items (failed ones aren't
+     * billed) plus `usage.generated_images`, the count BytePlus bills.
+     */
+    private requestSeedream;
     getCreditUsed(params: CreditUsageParams): number;
 }
 //# sourceMappingURL=bytedance.service.d.ts.map

package/lib/services/aiGen/providers/bytedance/bytedance.service.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"bytedance.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/bytedance/bytedance.service.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAWlB,qBAAa,gBAAiB,SAAQ,wBAAwB;IAC5D,OAAO,CAAC,QAAQ,CAAC,OAAO,CACsD;IAI9E,OAAO,CAAC,cAAc;IAYtB,OAAO,CAAC,WAAW;IAIb,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA2M3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAqIjD,aAAa,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM;CAyEjD"}
+{"version":3,"file":"bytedance.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/bytedance/bytedance.service.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AA6BlB;;;;;;GAMG;AACH,wBAAgB,YAAY,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAQlF;AAsBD,qBAAa,gBAAiB,SAAQ,wBAAwB;IAC5D,OAAO,CAAC,QAAQ,CAAC,OAAO,CACsD;IAQ9E,OAAO,CAAC,cAAc;IAYtB,OAAO,CAAC,WAAW;IAIb,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAoP3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAqIjD;;;;;;;;;;;OAWG;IACG,aAAa,CAAC,MAAM,EAAE,qBAAqB,GAAG,OAAO,CAAC,qBAAqB,CAAC;IAyIlF,+EAA+E;YACjE,mBAAmB;IAoCjC;;;OAGG;YACW,eAAe;IAgD7B,aAAa,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM;CAsFjD"}