@shop-os/modelkit 0.2.15 → 0.2.16

package/dist/index.cjs

@@ -1512,7 +1512,14 @@ var VIDEO_MODELS = [
       durations: [3, 4, 5, 6, 7, 8, 9, 10],
       references: { images: 6, videos: 1, videoSeconds: { max: 10 } },
       sourceVideoSeconds: { max: 10 },
-      supports: []
+      // Extension (probe 2026-09-24, 360p): each `taskType: "extend"` call
+      // renders `durationSeconds` more and returns the FULL cumulative clip
+      // (3 s -> 6 s -> 9 s); the provider caps the clip at 40 s. A first
+      // extension can upload a <= 10 s source with task "extend"; later ones
+      // chain by `previousInteractionId` (the API rejects the id together
+      // with a video task, so those calls carry the prompt alone).
+      supports: ["videoExtension"],
+      extension: { segmentSeconds: 10, maxTotalSeconds: 40 }
     },
     defaults: {},
     usage: { ...VIDEO_USAGE },
@@ -1520,7 +1527,7 @@ var VIDEO_MODELS = [
     display: {
       provider: "google",
       logo: GOOGLE_LOGO2,
-      description: "Google Gemini Omni Flash 1.1 conversational video generation and editing; text, image, and up to 10-second video input, 3-10 second output at up to 4K and 24 FPS."
+      description: "Google Gemini Omni Flash 1.1 conversational video generation, editing and extension; text, image, and up to 10-second video input, 3-10 second output at up to 4K and 24 FPS, extendable to 40 seconds."
     },
     capabilities: { acceptsSourceVideo: true }
   },
@@ -2254,6 +2261,7 @@ var SupportedOptionSchema = import_zod2.z.enum([
   "watermark",
   "safetyChecker",
   "taskType",
+  "videoExtension",
   "instrumental"
 ]);
 var UsageUnitSchema = import_zod2.z.enum([
@@ -2301,7 +2309,8 @@ var ModelConstraintsSchema = import_zod2.z.looseObject({
   requires: import_zod2.z.array(RequiredInputSchema).optional(),
   supports: import_zod2.z.array(SupportedOptionSchema).optional(),
   ratioIgnoredWith: import_zod2.z.array(import_zod2.z.enum(["firstFrame", "sourceVideo"])).optional(),
-  sourceVideoSeconds: VideoSecondsLimitsSchema.omit({ total: true }).extend({ inReferenceTotal: import_zod2.z.boolean().optional() }).optional()
+  sourceVideoSeconds: VideoSecondsLimitsSchema.omit({ total: true }).extend({ inReferenceTotal: import_zod2.z.boolean().optional() }).optional(),
+  extension: import_zod2.z.looseObject({ segmentSeconds: import_zod2.z.number(), maxTotalSeconds: import_zod2.z.number() }).optional()
 });
 var MediaRefSchema = import_zod2.z.union([
   import_zod2.z.looseObject({
@@ -4568,12 +4577,34 @@ async function generateOmniVideo(req, ctx, createGenAIClient) {
   const modelId = ctx.model.id;
   const providerModelId = ctx.model.providerModelId;
   const duration = req.durationSeconds ?? 10;
-  const images = req.referenceImages ?? [];
-  const videos = req.referenceVideos ?? [];
-  const task = req.sourceVideo ? "edit" : req.firstFrame ? "image_to_video" : images.length > 0 || videos.length > 0 ? "reference_to_video" : "text_to_video";
+  const extend = req.taskType === "extend";
+  const previousId = extend ? req.previousInteractionId : void 0;
+  const chained = previousId !== void 0;
+  for (const field of ["referenceImages", "referenceVideos", "firstFrame"]) {
+    const value = req[field];
+    if (!chained || value === void 0 || Array.isArray(value) && value.length === 0)
+      continue;
+    ctx.warnings.push({
+      code: "option_ignored",
+      field,
+      from: Array.isArray(value) ? value.length : "set",
+      message: `${modelId} chained extension sends the prompt alone; ${field} ignored`
+    });
+  }
+  const images = chained ? [] : req.referenceImages ?? [];
+  const videos = chained ? [] : req.referenceVideos ?? [];
+  const firstFrame = chained ? void 0 : req.firstFrame;
+  if (extend && !chained && req.sourceVideo === void 0) {
+    throw new ModelKitError({
+      code: "missing_input",
+      message: `${modelId} extend needs a source video or a previousInteractionId`,
+      model: modelId
+    });
+  }
+  const task = extend ? "extend" : req.sourceVideo ? "edit" : firstFrame ? "image_to_video" : images.length > 0 || videos.length > 0 ? "reference_to_video" : "text_to_video";
   const isOmniV1 = providerModelId === "gemini-omni-flash-preview";
   const responseFormat = { type: "video", delivery: "uri" };
-  if (req.sourceVideo === void 0 && req.aspectRatio !== void 0 && req.aspectRatio !== "adaptive") {
+  if (req.sourceVideo === void 0 && !extend && req.aspectRatio !== void 0 && req.aspectRatio !== "adaptive") {
     responseFormat.aspect_ratio = toOmniAspect(req.aspectRatio);
   }
   if (!isOmniV1) {
@@ -4605,18 +4636,26 @@ async function generateOmniVideo(req, ctx, createGenAIClient) {
   const cleanupUploads = async () => {
     for (const fileName of uploadedNames) await deleteGoogleFile(fileName, apiKey, ctx.fetch);
   };
-  const callCreate = (input) => client2.interactions.create({
-    model: providerModelId,
-    input,
-    response_format: responseFormat,
-    generation_config: { video_config: { task } },
-    store: true
-  });
+  const callCreate = (input) => client2.interactions.create(
+    previousId !== void 0 ? {
+      model: providerModelId,
+      input,
+      response_format: responseFormat,
+      store: true,
+      previous_interaction_id: previousId
+    } : {
+      model: providerModelId,
+      input,
+      response_format: responseFormat,
+      generation_config: { video_config: { task } },
+      store: true
+    }
+  );
   const createRetryOptions = mediaRetryOptions(ctx.call, ctx.signal);
   let interaction;
   try {
     const parts = [];
-    if (req.sourceVideo) {
+    if (req.sourceVideo && !chained) {
       const media = await resolveMedia(
         req.sourceVideo,
         ctx.fetch,
@@ -4639,11 +4678,11 @@ async function generateOmniVideo(req, ctx, createGenAIClient) {
         mime_type: media.mime
       });
     }
-    if (req.firstFrame) {
-      if (req.firstFrame.label !== void 0)
-        parts.push({ type: "text", text: req.firstFrame.label });
+    if (firstFrame) {
+      if (firstFrame.label !== void 0)
+        parts.push({ type: "text", text: firstFrame.label });
       const media = await resolveMedia(
-        req.firstFrame,
+        firstFrame,
         ctx.fetch,
         ctx.signal,
         "image/png",
@@ -4746,6 +4785,7 @@ async function generateOmniVideo(req, ctx, createGenAIClient) {
         return void 0;
       }, pollOptions);
       const video = extractOmniVideo(done, modelId);
+      const providerRef = { interactionId: done.id ?? jobId };
       if (video.data) {
         return {
           video: {
@@ -4753,14 +4793,16 @@ async function generateOmniVideo(req, ctx, createGenAIClient) {
             mime: video.mime_type ?? "video/mp4"
           },
           usage: usage2,
-          raw: done
+          raw: done,
+          providerRef
         };
       }
       if (video.uri)
         return {
           video: await downloadVideo(ctx, video.uri, apiKey, signal),
           usage: usage2,
-          raw: done
+          raw: done,
+          providerRef
         };
       throw classifyHttpError(
         void 0,
@@ -7652,6 +7694,7 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
   setOrDelete(merged, "referenceImages", images);
   setOrDelete(merged, "referenceVideos", videos);
   setOrDelete(merged, "referenceAudios", audios);
+  const keepExtend = merged.taskType === "extend" && (model.constraints.supports ?? []).includes("videoExtension");
   dropUnsupported(merged, model, warnings, [
     "seed",
     "negativePrompt",
@@ -7661,8 +7704,17 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
     "lastFrame",
     "watermark",
     "safetyChecker",
-    "taskType"
+    ...keepExtend ? [] : ["taskType"]
   ]);
+  if (merged.previousInteractionId !== void 0 && !keepExtend) {
+    warnings.push({
+      code: "option_ignored",
+      field: "previousInteractionId",
+      from: merged.previousInteractionId,
+      message: `${model.id} does not support previousInteractionId; ignored`
+    });
+    delete merged.previousInteractionId;
+  }
   return { request: merged, warnings };
 }
 function resolveAudioOptions(model, request, onUnsupported = "error") {