vidspotai-shared 1.0.79 → 1.0.81

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

@@ -13,32 +13,83 @@ const helpers_1 = require("./helpers");
 const helpers_2 = require("../../helpers");
 const utils_1 = require("../../../../utils");
 const logger_1 = require("../../../../utils/logger");
-// DashScope hosts Wan endpoints under three task-typed paths. We pick at submit
-// time based on which inputs are present (audio → s2v, image → i2v, else t2v).
-const ALIBABA_BASE = "https://dashscope-intl.aliyuncs.com/api/v1/services/aigc";
-const ENDPOINT_T2V = `${ALIBABA_BASE}/video-generation/video-synthesis`;
-const ENDPOINT_I2V = `${ALIBABA_BASE}/image2video/video-synthesis`;
-const ENDPOINT_S2V = `${ALIBABA_BASE}/sound2video/video-synthesis`;
-// All three share the same /tasks/{task_id} status endpoint.
-const ENDPOINT_TASK_STATUS = "https://dashscope-intl.aliyuncs.com/api/v1/tasks";
+const errors_1 = require("../../../../utils/errors");
+// Endpoint topology (verified 2026-06-06 against intl docs + live probes):
+//
+//   T2V + I2V        → /services/aigc/video-generation/video-synthesis   (async; X-DashScope-Async: enable)
+//   Wan/Qwen async   → /services/aigc/text2image/image-synthesis         (async; legacy qwen-image, qwen-image-plus)
+//   Wan async image  → /services/aigc/image-generation/generation        (async; wan2.7-image*, wan2.6-image)
+//   Qwen/Wan SYNC    → /services/aigc/multimodal-generation/generation   (sync; new qwen-image-2.0*, qwen-image-edit*, wan2.7-image* + edit)
+//
+// Both T2V and I2V use the SAME video endpoint — the model + presence of
+// img_url determines mode. The legacy split into /image2video/... and
+// /sound2video/... endpoints was wrong (Model not exist errors).
+//
+// S2V (wan2.2-s2v) is China-region only and cannot be reached with an intl
+// DashScope key; the model key is intentionally kept but rejected at submit
+// with a UserFacingError. Same for kolors-v2 (Kuaishou model, not Alibaba).
+const ALIBABA_BASE_URL = "https://dashscope-intl.aliyuncs.com/api/v1";
+const ALIBABA_BASE = `${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`;
+const ENDPOINT_IMAGE_GEN_ASYNC = `${ALIBABA_BASE}/image-generation/generation`;
+const ENDPOINT_MULTIMODAL_SYNC = `${ALIBABA_BASE}/multimodal-generation/generation`;
+const ENDPOINT_TASK_STATUS = `${ALIBABA_BASE_URL}/tasks`;
+// Model IDs that don't run on the intl `dashscope-intl.aliyuncs.com` endpoint.
+// We surface a clean PROVIDER_AUTH_ERROR rather than letting them fall through
+// and hit a 404 "Model not exist."
+const INTL_UNAVAILABLE_MODEL_IDS = new Set([
+    "wan2.2-s2v", // China region only (dashscope.aliyuncs.com)
+    "kolors-v2", // Kuaishou model; not hosted on DashScope intl
+]);
+// Image model classification — drives sync vs async endpoint dispatch.
+// Sync (multimodal-generation/generation, returns inline image URL):
+//   - Qwen 2.x sync line: qwen-image-2.0, qwen-image-2.0-pro, qwen-image-max,
+//     qwen-image-edit*, plus wan2.7-image / wan2.7-image-pro when used inline.
+// Async-via-image-generation (returns task_id, poll /tasks):
+//   - Wan image-gen line: wan2.7-image*, wan2.6-image.
+// Async-via-text2image (returns task_id, poll /tasks):
+//   - Legacy qwen-image, qwen-image-plus.
+function isSyncMultimodalImageModel(modelId) {
+    return (modelId.startsWith("qwen-image-2") ||
+        modelId.startsWith("qwen-image-max") ||
+        modelId.startsWith("qwen-image-edit") ||
+        modelId.startsWith("wan2.7-image"));
+}
+function isAsyncWanImageModel(modelId) {
+    return modelId.startsWith("wan2.7-image") || modelId.startsWith("wan2.6-image");
+}
 class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderService {
     constructor() {
         super();
-        this.baseUrl = ENDPOINT_T2V; // legacy default; used by getCreditUsed only
+        // Used by getCreditUsed only. Real request paths are chosen per-call.
+        this.baseUrl = ENDPOINT_VIDEO_ASYNC;
         this.timeout = 60000; // 60 seconds
         if (!process.env.ALIBABA_API_KEY) {
-            throw new Error("Missing ALIBABA_API_KEY in environment variables");
+            // Classify as PROVIDER_AUTH_ERROR (UserFacingError → warn log, no Slack
+            // page per job). The DashScope key is single-Bearer and distinct from
+            // ALIBABA_CLOUD_ACCESS_KEY/SECRET (those are general Alibaba Cloud
+            // creds, not DashScope) — surface a hint so an operator knows what to
+            // provision instead of chasing a generic 500.
+            throw new errors_1.UserFacingError("Alibaba (DashScope) API key is not configured. Set ALIBABA_API_KEY (Bearer sk-* token from dashscope.aliyuncs.com) in the runtime env.", errors_1.USER_FACING_ERROR_CODES.PROVIDER_AUTH_ERROR);
         }
     }
-    async request(body, method = "POST", url = this.baseUrl) {
+    /**
+     * DashScope API call. `async` toggles the `X-DashScope-Async: enable` header
+     * — required for async endpoints (video-generation, image-generation,
+     * text2image), MUST be omitted for sync multimodal-generation.
+     */
+    async request(body, method = "POST", url = this.baseUrl, asyncMode = true) {
+        const headers = {
+            Authorization: `Bearer ${process.env.ALIBABA_API_KEY}`,
+            "Content-Type": "application/json",
+        };
+        if (asyncMode)
+            headers["X-DashScope-Async"] = "enable";
         const config = {
             method,
             url,
-            headers: {
-                Authorization: `Bearer ${process.env.ALIBABA_API_KEY}`,
-                "Content-Type": "application/json",
-                "X-DashScope-Async": "enable",
-            },
+            headers,
             timeout: this.timeout,
             data: method === "POST" ? body : undefined,
         };
@@ -51,10 +102,12 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
         const modelId = modelConfig?.modelId;
         if (!modelId)
             throw new Error(`Unknown modelKey: ${params.modelKey}`);
-        // Endpoint selection: audio wins (S2V), then image (I2V), else T2V.
-        const isS2V = !!params.inputAudioUrl;
-        const isI2V = !isS2V && !!params.inputImageUrl;
-        const endpoint = isS2V ? ENDPOINT_S2V : isI2V ? ENDPOINT_I2V : ENDPOINT_T2V;
+        if (INTL_UNAVAILABLE_MODEL_IDS.has(modelId)) {
+            throw new errors_1.UserFacingError(`Model "${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;
         const input = {};
         if (params.prompt)
             input.prompt = params.prompt;
@@ -62,21 +115,16 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
             input.negative_prompt = params.negativePrompt;
         if (isI2V) {
             input.img_url = params.inputImageUrl;
-            // Wan 2.7 i2v supports first-last-frame interpolation
+            // wan2.7-i2v supports first-last-frame interpolation
             if (params.lastFrameImageUrl)
                 input.last_frame_url = params.lastFrameImageUrl;
         }
-        if (isS2V) {
-            input.audio_url = params.inputAudioUrl;
-            if (params.inputImageUrl)
-                input.image_url = params.inputImageUrl;
-        }
         const parameters = {
             duration: params.duration || 5,
             prompt_extend: params.promptOptimizer ?? true,
         };
-        // T2V requires explicit size; I2V/S2V derive from inputs.
-        if (!isI2V && !isS2V) {
+        // T2V requires explicit size; I2V derives dimensions from the input image.
+        if (!isI2V) {
             const size = (0, helpers_1.getAlibabaDimensions)(params.resolution, params.aspectRatio);
             if (!size) {
                 throw new Error(`Invalid resolution/aspect ratio combination: ${params.resolution} ${params.aspectRatio}`);
@@ -88,7 +136,7 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
         if (params.watermark !== undefined)
             parameters.watermark = params.watermark;
         const body = { model: modelId, input, parameters };
-        const result = await this.request(body, "POST", endpoint);
+        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;
         if (!result || !taskId) {
@@ -161,38 +209,127 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
         return { status: types_1.EVideoSceneStatus.PENDING };
     }
     /**
-     * DashScope image generation. Both Kolors and Qwen-Image are async on
-     * DashScope — submit returns a task_id, then poll /tasks/{id}. Image jobs
-     * typically finish in 5–15s, well inside our HTTP timeout, so we poll
-     * inline rather than going through BullMQ.
+     * DashScope image generation. Three dispatch paths (verified live 2026-06-06):
+     *
+     *   1. SYNC multimodal-generation/generation
+     *      Models: qwen-image-2.0, qwen-image-2.0-pro, qwen-image-max,
+     *              qwen-image-edit*, wan2.7-image, wan2.7-image-pro
+     *      Response: output.choices[0].message.content[].image (inline URLs)
+     *      Note: NO X-DashScope-Async header. Edit mode is triggered by including
+     *            {image: refUrl} entries in the user message content array.
+     *
+     *   2. ASYNC image-generation/generation (Wan async image)
+     *      Models: wan2.7-image*, wan2.6-image (when caller prefers async; we
+     *              currently route these through path 1 since they're available
+     *              there too).
      *
-     * Endpoint selection:
-     *   - Qwen-Image edit (modelId qwen-image-edit*): /image2image/image-synthesis
-     *   - Everything else (Qwen T2I, Kolors): /text2image/image-synthesis
+     *   3. ASYNC text2image/image-synthesis (legacy)
+     *      Models: qwen-image, qwen-image-plus
+     *      Response: output.results[].url
+     *
+     * Image jobs typically finish in 5–15s, well inside our HTTP timeout, so
+     * we poll inline rather than going through BullMQ.
      */
     async generateImage(params) {
         const modelConfig = aiModels_1.aiModelConfigs[params.modelKey];
         const modelId = modelConfig?.modelId;
         if (!modelId)
             throw new Error(`Unknown image modelKey: ${params.modelKey}`);
-        const isEdit = modelId.startsWith("qwen-image-edit");
-        const submitUrl = isEdit
-            ? `${ALIBABA_BASE}/image2image/image-synthesis`
-            : `${ALIBABA_BASE}/text2image/image-synthesis`;
+        if (INTL_UNAVAILABLE_MODEL_IDS.has(modelId)) {
+            throw new errors_1.UserFacingError(`Model "${modelId}" is not available on the intl DashScope endpoint. Use a different image provider.`, errors_1.USER_FACING_ERROR_CODES.PROVIDER_AUTH_ERROR);
+        }
+        const refs = [
+            ...(params.inputImageUrl ? [params.inputImageUrl] : []),
+            ...(params.inputImageUrls ?? []),
+        ];
+        const isEdit = modelId.startsWith("qwen-image-edit") ||
+            (refs.length > 0 && (modelId.startsWith("wan2.7-image") || modelId.startsWith("qwen-image-2")));
+        if (isEdit && !refs.length) {
+            throw new Error(`${modelId}: edit/reference mode requires at least one input image`);
+        }
+        let remoteUrls;
+        let providerRequestId;
+        if (isSyncMultimodalImageModel(modelId)) {
+            const result = await this.generateImageSync(modelId, params, refs);
+            remoteUrls = result.urls;
+            providerRequestId = result.requestId;
+        }
+        else if (isAsyncWanImageModel(modelId)) {
+            const result = await this.generateImageAsync(modelId, params, ENDPOINT_IMAGE_GEN_ASYNC);
+            remoteUrls = result.urls;
+            providerRequestId = result.taskId;
+        }
+        else {
+            // Legacy async path: qwen-image, qwen-image-plus.
+            const result = await this.generateImageAsync(modelId, params, ENDPOINT_IMAGE_T2I_ASYNC);
+            remoteUrls = result.urls;
+            providerRequestId = result.taskId;
+        }
+        if (!remoteUrls.length) {
+            throw new Error("DashScope image returned no URLs");
+        }
+        const bucket = (0, firebase_1.getBucket)();
+        const ts = Date.now();
+        const urls = [];
+        for (let i = 0; i < remoteUrls.length; i++) {
+            const remoteUrl = remoteUrls[i];
+            const bytes = Buffer.from(await (await axios_1.default.get(remoteUrl, { responseType: "arraybuffer", timeout: this.timeout })).data);
+            const path = `images/dashscope/${ts}-${Math.random().toString(36).slice(2, 8)}-${i}.png`;
+            const file = bucket.file(path);
+            await file.save(bytes, { contentType: "image/png" });
+            const [signed] = await file.getSignedUrl({ action: "read", expires: "03-09-2491" });
+            urls.push(signed);
+        }
+        return {
+            imageUrl: urls[0],
+            imageUrls: urls.length > 1 ? urls : undefined,
+            providerRequestId,
+        };
+    }
+    /**
+     * Sync multimodal-generation. Returns inline image URLs in
+     * `output.choices[0].message.content[].image`. No polling needed.
+     */
+    async generateImageSync(modelId, params, refs) {
+        const userContent = [];
+        // Image refs come first (DashScope convention); each as its own content entry.
+        for (const ref of refs.slice(0, 3)) {
+            userContent.push({ image: ref });
+        }
+        if (params.prompt)
+            userContent.push({ text: params.prompt });
+        const input = {
+            messages: [{ role: "user", content: userContent }],
+        };
+        const parameters = {};
+        if (params.negativePrompt)
+            parameters.negative_prompt = params.negativePrompt;
+        if (params.watermark !== undefined)
+            parameters.watermark = params.watermark;
+        if (params.promptOptimizer !== undefined)
+            parameters.prompt_extend = params.promptOptimizer;
+        if (params.seed !== undefined)
+            parameters.seed = params.seed;
+        if (params.aspectRatio)
+            parameters.size = params.aspectRatio;
+        if (params.imageSize)
+            parameters.size = params.imageSize;
+        const result = await this.request({ model: modelId, input, parameters }, "POST", ENDPOINT_MULTIMODAL_SYNC, false);
+        const choices = result?.output?.choices ?? [];
+        const content = choices[0]?.message?.content ?? [];
+        const urls = content
+            .map((c) => c?.image)
+            .filter((u) => typeof u === "string" && !!u);
+        return { urls, requestId: result?.request_id ?? "" };
+    }
+    /**
+     * Async submit + poll. Used by both image-generation/generation (Wan async)
+     * and text2image/image-synthesis (legacy Qwen).
+     */
+    async generateImageAsync(modelId, params, submitUrl) {
         const input = { prompt: params.prompt };
         if (params.negativePrompt)
             input.negative_prompt = params.negativePrompt;
-        if (isEdit) {
-            const refs = [
-                ...(params.inputImageUrl ? [params.inputImageUrl] : []),
-                ...(params.inputImageUrls ?? []),
-            ];
-            if (!refs.length) {
-                throw new Error("Qwen-Image edit requires at least one input image");
-            }
-            // Qwen edit accepts 1–3 refs as base_image_url.
-            input.base_image_url = refs.length === 1 ? refs[0] : refs.slice(0, 3);
-        }
         const parameters = {
             n: params.numImages ?? 1,
             ...(params.aspectRatio ? { size: params.aspectRatio } : {}),
@@ -203,18 +340,18 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
                 ? { prompt_extend: params.promptOptimizer }
                 : { prompt_extend: true }),
         };
-        const submit = await this.request({ model: modelId, input, parameters }, "POST", submitUrl);
+        const submit = await this.request({ model: modelId, input, parameters }, "POST", submitUrl, true);
         const taskId = submit?.output?.task_id;
         if (!taskId) {
             throw new Error("DashScope image submit returned no task_id");
         }
-        // Poll up to 90s, every 2s. Image jobs are quick.
+        // Poll up to 90s, every 2s.
         const maxAttempts = 45;
         const intervalMs = 2000;
         let lastResult = null;
         for (let i = 0; i < maxAttempts; i++) {
             await new Promise((r) => setTimeout(r, intervalMs));
-            const poll = await this.request(null, "GET", `${ENDPOINT_TASK_STATUS}/${taskId}`);
+            const poll = await this.request(null, "GET", `${ENDPOINT_TASK_STATUS}/${taskId}`, false);
             const status = poll?.output?.task_status ?? poll?.status;
             if (status === "SUCCEEDED" || status === "succeeded") {
                 lastResult = poll;
@@ -228,27 +365,8 @@ class AlibabaService extends baseAiGenProvider_service_1.BaseAiGenProviderServic
             throw new Error("DashScope image task timed out after 90s");
         }
         const results = lastResult.output?.results ?? [];
-        const remoteUrls = results.map((r) => r.url).filter((u) => !!u);
-        if (!remoteUrls.length) {
-            throw new Error("DashScope image task returned no URLs");
-        }
-        const bucket = (0, firebase_1.getBucket)();
-        const ts = Date.now();
-        const urls = [];
-        for (let i = 0; i < remoteUrls.length; i++) {
-            const remoteUrl = remoteUrls[i];
-            const bytes = Buffer.from(await (await axios_1.default.get(remoteUrl, { responseType: "arraybuffer", timeout: this.timeout })).data);
-            const path = `images/dashscope/${ts}-${Math.random().toString(36).slice(2, 8)}-${i}.png`;
-            const file = bucket.file(path);
-            await file.save(bytes, { contentType: "image/png" });
-            const [signed] = await file.getSignedUrl({ action: "read", expires: "03-09-2491" });
-            urls.push(signed);
-        }
-        return {
-            imageUrl: urls[0],
-            imageUrls: urls.length > 1 ? urls : undefined,
-            providerRequestId: taskId,
-        };
+        const urls = results.map((r) => r.url).filter((u) => !!u);
+        return { urls, taskId };
     }
     getCreditUsed({ modelKey, resolution, aspectRatio, duration, multiClip = false, numImages = 1, }) {
         const modelConfig = aiModels_1.aiModelConfigs[modelKey];

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

@@ -13,6 +13,7 @@ export declare class GoogleService extends BaseAiGenProviderService {
     generateVideo(params: VideoGenerationParams): Promise<VideoGenerationResult>;
     checkVideoStatus({ task, outputFilename, outputFilePath, }: VideoStatusParams): Promise<VideoStatusResult>;
     generateImage(params: ImageGenerationParams): Promise<ImageGenerationResult>;
+    private _generateImage;
     /**
      * Lyria 2 (Vertex AI). Sync — POST {region}-aiplatform.googleapis.com/.../lyria-002:predict
      * returns base64-encoded WAV audio inline. Auth via ADC on the function service

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

@@ -1 +1 @@
-{"version":3,"file":"google.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/google/google.service.ts"],"names":[],"mappings":"AAgBA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AA6IlB,qBAAa,aAAc,SAAQ,wBAAwB;IACzD,OAAO,CAAC,EAAE,CAAc;IACxB,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,kBAAkB,CAAK;;IAO/C;;;;OAIG;YACW,kBAAkB;IA+B1B,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAqF3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IA0F3C,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAsGjC;;;;;;OAMG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAqEjC,aAAa,CAAC,EAAE,QAAQ,EAAE,QAAY,EAAE,UAAmB,EAAE,SAAiB,EAAE,SAAa,EAAE,SAAS,EAAE,EAAE,iBAAiB,GAAG,MAAM;CA8BvI"}
+{"version":3,"file":"google.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/google/google.service.ts"],"names":[],"mappings":"AAgBA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAmKlB,qBAAa,aAAc,SAAQ,wBAAwB;IACzD,OAAO,CAAC,EAAE,CAAc;IACxB,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,kBAAkB,CAAK;;IAO/C;;;;OAIG;YACW,kBAAkB;IA+B1B,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAyG3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IA0F3C,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;YAiBnB,cAAc;IAwG5B;;;;;;OAMG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAqEjC,aAAa,CAAC,EAAE,QAAQ,EAAE,QAAY,EAAE,UAAmB,EAAE,SAAiB,EAAE,SAAa,EAAE,SAAS,EAAE,EAAE,iBAAiB,GAAG,MAAM;CA8BvI"}

package/lib/services/aiGen/providers/google/google.service.js

@@ -127,15 +127,30 @@ function classifyGoogleApiError(err) {
         if (httpCode === 14 || /high demand/i.test(msg)) {
             return new errors_1.UserFacingError(msg, errors_1.USER_FACING_ERROR_CODES.VIDEO_PROVIDER_HIGH_DEMAND);
         }
-        // INVALID_ARGUMENT 400 — typically "Your use case is currently not
-        // supported" when the request mixes features the chosen Veo variant
-        // doesn't offer (e.g. referenceImages on -fast, lastFrame on -lite,
-        // video-extension with a non-Veo source URL). Not actionable as a
-        // platform bug — surface as a translatable capability mismatch so the
-        // user can switch model/feature combo instead of firing Slack.
-        if (status === "INVALID_ARGUMENT" || httpCode === 400) {
+        // INVALID_ARGUMENT 400 — narrow match: only the specific "use case is
+        // currently not supported" string, which Veo returns when our request
+        // structure doesn't match the chosen model variant's capabilities.
+        // The pre-call guards above (duration=8 for lastFrame/refs) should
+        // prevent the known cases; if we still hit this it's a NEW combo we
+        // haven't profiled — surface as CAPABILITY_MISMATCH so the user gets a
+        // useful message, AND keep the raw provider text in the error so the
+        // next entry in PROD_FIX_LOG can identify which combo broke. Generic
+        // 400s (other INVALID_ARGUMENT variants) still surface as `error` so
+        // a real platform bug isn't muted.
+        if ((status === "INVALID_ARGUMENT" || httpCode === 400) &&
+            /use case is currently not supported/i.test(msg)) {
             return new errors_1.UserFacingError(msg, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
         }
+        // Imagen + Nano-Banana surface Responsible-AI filter rejections as
+        // INVALID_ARGUMENT 400 with the literal text "filtered out because they
+        // violated Google's Responsible AI practices" (and a recommendation to
+        // rephrase). This is user content moderation, not a system bug — show the
+        // user the rephrase hint and skip the Slack page. Also matches the Veo
+        // RAI message ("violated Google's content policies") for the same reason.
+        if ((status === "INVALID_ARGUMENT" || httpCode === 400) &&
+            /(filtered out because they violated|violated Google's (?:Responsible AI|content) (?:practices|policies))/i.test(msg)) {
+            return new errors_1.UserFacingError("Your prompt was flagged by Google's safety filters. Please rephrase and try again.", errors_1.USER_FACING_ERROR_CODES.CONTENT_POLICY_VIOLATION);
+        }
     }
     catch {
         // Not JSON — fall through to non-JSON checks.
@@ -192,6 +207,21 @@ class GoogleService extends baseAiGenProvider_service_1.BaseAiGenProviderService
         const modelConfig = aiModels_1.aiModelConfigs[params.modelKey];
         const modelId = modelConfig.modelId;
         const isVeo3_1 = VEO_3_1_MODELS.has(params.modelKey);
+        // Gemini Veo cross-constraint: first+last-frame interpolation AND
+        // reference images BOTH require durationSeconds=8. Sending any other
+        // duration returns INVALID_ARGUMENT 400 "Your use case is currently not
+        // supported." with no hint about which param caused it. Surface a
+        // typed UserFacingError so the user/frontend can correct the input
+        // instead of burning a provider call + opaque rejection.
+        const needsDuration8 = !!params.lastFrameImageUrl ||
+            (isVeo3_1 && (params.referenceImageUrls?.length ?? 0) > 0);
+        if (needsDuration8 && params.duration !== undefined && params.duration !== 8) {
+            const constraint = params.lastFrameImageUrl
+                ? "first-frame + last-frame interpolation"
+                : "reference images";
+            throw new errors_1.UserFacingError(`Google Veo requires an 8-second duration when using ${constraint}. ` +
+                `Please select 8s or remove the ${params.lastFrameImageUrl ? "last-frame image" : "reference images"}.`, errors_1.USER_FACING_ERROR_CODES.CAPABILITY_MISMATCH);
+        }
         const request = {
             model: modelId,
             prompt: params.prompt,
@@ -318,6 +348,24 @@ class GoogleService extends baseAiGenProvider_service_1.BaseAiGenProviderService
         return { status: types_1.EVideoSceneStatus.PENDING };
     }
     async generateImage(params) {
+        try {
+            return await this._generateImage(params);
+        }
+        catch (err) {
+            // The SDK's generateImages / generateContent throws ApiError instances
+            // whose .message is a JSON string. Route through classifyGoogleApiError
+            // so Imagen RAI safety filter rejections (the most common failure mode
+            // for image gen — "filtered out because they violated Google's
+            // Responsible AI practices") become UserFacingError(CONTENT_POLICY_VIOLATION)
+            // instead of leaking as raw provider JSON into the worker's Slack
+            // error channel.
+            const userFacing = classifyGoogleApiError(err);
+            if (userFacing)
+                throw userFacing;
+            throw err;
+        }
+    }
+    async _generateImage(params) {
         const modelConfig = aiModels_1.aiModelConfigs[params.modelKey];
         const modelId = modelConfig?.modelId;
         if (!modelId)

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

@@ -1 +1 @@
-{"version":3,"file":"openai.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/openai/openai.service.ts"],"names":[],"mappings":"AAiBA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,oBAAoB,EACpB,oBAAoB,EACpB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAElB,qBAAa,aAAc,SAAQ,wBAAwB;IACzD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAa;IACxC,OAAO,CAAC,MAAM,CAAS;;IAQjB,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAgE3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAsC3C,YAAY,CAChB,MAAM,EAAE,oBAAoB,GAC3B,OAAO,CAAC,oBAAoB,CAAC;IA2ChC;;;;OAIG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA6HjC,aAAa,CAAC,EACZ,QAAQ,EACR,UAAuB,EACvB,QAAY,EACZ,SAAiB,EACjB,SAAa,EACb,OAAO,GACR,EAAE,iBAAiB,GAAG,MAAM;CAoB9B"}
+{"version":3,"file":"openai.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/openai/openai.service.ts"],"names":[],"mappings":"AAiBA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,oBAAoB,EACpB,oBAAoB,EACpB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AAElB,qBAAa,aAAc,SAAQ,wBAAwB;IACzD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAa;IACxC,OAAO,CAAC,MAAM,CAAS;;IAQjB,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA0E3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAsC3C,YAAY,CAChB,MAAM,EAAE,oBAAoB,GAC3B,OAAO,CAAC,oBAAoB,CAAC;IA2ChC;;;;OAIG;IACG,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IA6HjC,aAAa,CAAC,EACZ,QAAQ,EACR,UAAuB,EACvB,QAAY,EACZ,SAAiB,EACjB,SAAa,EACb,OAAO,GACR,EAAE,iBAAiB,GAAG,MAAM;CAoB9B"}

package/lib/services/aiGen/providers/openai/openai.service.js

@@ -61,17 +61,29 @@ class OpenaiService extends baseAiGenProvider_service_1.BaseAiGenProviderService
         };
         // First-frame image-to-video. The Sora API requires the reference image
         // dimensions to match `size`; the caller is responsible for that.
-        // Note: openai SDK v6.3.0 does NOT yet expose extensions / characters /
-        // edits endpoints — those exist in the REST API but are not surfaced here.
+        //
+        // We use the JSON variant `input_reference: { image_url }` (OpenAI
+        // fetches the image themselves) rather than the multipart `Uploadable`
+        // shape. Reason: as of late 2026 sora-2-pro started returning
+        //   400 Invalid type for 'input_reference': expected an object, but
+        //       got a file instead.
+        // for multipart uploads that worked on sora-2. The JSON form is
+        // documented for both models and avoids the size mismatch entirely.
+        //
+        // openai SDK 6.3.0's typed shape for input_reference is `Uploadable`
+        // (no JSON-object overload yet), so we bypass the type with a cast.
+        // The HTTP layer serializes plain objects as JSON automatically when
+        // no Uploadable is present in the body, which is exactly what we want.
+        //
+        // Note: openai SDK v6.3.0 does NOT yet expose extensions / characters
+        // / edits endpoints — those exist in the REST API but are not surfaced.
         if (params.inputImageUrl) {
-            const resp = await fetch(params.inputImageUrl);
-            // fetch() does not throw on 4xx/5xx — surface the HTTP error so we
-            // don't ship an HTML error page to Sora as if it were image bytes.
-            if (!resp.ok) {
-                throw new errors_1.UserFacingError(`Input image could not be downloaded (HTTP ${resp.status}). The image URL may have expired or been deleted.`);
-            }
-            const filename = (params.inputImageUrl.split("?")[0] ?? "reference").split("/").pop() || "reference.png";
-            request.input_reference = await (0, openai_1.toFile)(resp, filename);
+            // SDK 6.3.0 types input_reference as Uploadable only; the JSON
+            // object form is a runtime-supported overload that the types
+            // haven't caught up to yet. Bypass via unknown cast.
+            request.input_reference = {
+                image_url: params.inputImageUrl,
+            };
         }
         let job;
         try {

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

@@ -1 +1 @@
-{"version":3,"file":"pixverse.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/pixverse/pixverse.service.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AASlB,qBAAa,eAAgB,SAAQ,wBAAwB;IAC3D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAkD;IAKpE,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAsL3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAmGjD,aAAa,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM;CAkDjD"}
+{"version":3,"file":"pixverse.service.d.ts","sourceRoot":"","sources":["../../../../../src/services/aiGen/providers/pixverse/pixverse.service.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,wBAAwB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,EACL,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,iBAAiB,EACjB,iBAAiB,EAClB,MAAM,UAAU,CAAC;AA0FlB,qBAAa,eAAgB,SAAQ,wBAAwB;IAC3D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAkD;IAKpE,aAAa,CACjB,MAAM,EAAE,qBAAqB,GAC5B,OAAO,CAAC,qBAAqB,CAAC;IAmK3B,gBAAgB,CAAC,EACrB,IAAI,EACJ,cAAc,EACd,cAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;IA0FjD,aAAa,CAAC,MAAM,EAAE,iBAAiB,GAAG,MAAM;CAkDjD"}