@shop-os/modelkit 0.2.15 → 0.2.17

package/dist/catalog.d.cts

@@ -107,8 +107,14 @@ interface VideoRequest {
     watermark?: boolean;
     /** Provider-side safety checker toggle (fal MiniMax). Provider default when unset. */
     safetyChecker?: boolean;
-    /** Seedance omni-reference subtask hint. `edit` forces duration -1 and adaptive ratio. */
+    /** Seedance omni-reference subtask hint. `edit` forces duration -1 and adaptive ratio.
+     * `extend` on a `videoExtension` row continues the clip by `durationSeconds` more. */
     taskType?: "reference" | "edit" | "extend";
+    /** Continue a provider-side session instead of re-sending the clip (Gemini Omni
+     * Interactions `previous_interaction_id`). The prior call's `VideoResult.providerRef`
+     * carries it. The provider rejects it together with an explicit video task, so the
+     * adapter sends the prompt alone; `taskType: "extend"` still selects the extend path. */
+    previousInteractionId?: string;
     mode?: "standard" | "pro";
 }
 /**
@@ -160,7 +166,9 @@ interface MediaPart {
     cleanup?: () => Promise<void>;
 }
 type RequiredInput = "firstFrame" | "sourceVideo" | "references";
-type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "safetySettings" | "temperature" | "upscale" | "watermark" | "safetyChecker" | "taskType" | "instrumental";
+type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "safetySettings" | "temperature" | "upscale" | "watermark" | "safetyChecker" | "taskType" | "instrumental"
+/** Video rows that extend a clip in fixed segments up to a cumulative cap (see `ModelConstraints.extension`). */
+ | "videoExtension";
 interface VideoSecondsLimits {
     /** Minimum length of each reference video, in seconds. */
     min?: number;
@@ -210,6 +218,12 @@ interface ModelConstraints {
         /** The provider sends the source as a reference video, so it counts toward `references.videoSeconds.total` (Seedance). */
         inReferenceTotal?: boolean;
     };
+    /** Video extension (rows with `supports: ["videoExtension"]`): each `taskType: "extend"`
+     * call adds up to `segmentSeconds` and the clip can grow to `maxTotalSeconds`. */
+    extension?: {
+        segmentSeconds: number;
+        maxTotalSeconds: number;
+    };
 }
 /** JSON-compatible value. Provider options and raw payloads carry these, never `unknown`. */
 type JsonPrimitive = string | number | boolean | null;
@@ -351,6 +365,10 @@ interface VideoResult {
     };
     usage: UsageEvent;
     raw?: unknown;
+    /** Provider handle for chaining a follow-up call (`VideoRequest.previousInteractionId`). */
+    providerRef?: {
+        interactionId: string;
+    };
 }
 interface VideoCallJob extends VideoJob {
     warnings: Warning[];
@@ -613,6 +631,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             safetyChecker: "safetyChecker";
             taskType: "taskType";
             instrumental: "instrumental";
+            videoExtension: "videoExtension";
         }>>>;
         ratioIgnoredWith: z.ZodOptional<z.ZodArray<z.ZodEnum<{
             firstFrame: "firstFrame";
@@ -624,6 +643,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             editMin: z.ZodOptional<z.ZodNumber>;
             inReferenceTotal: z.ZodOptional<z.ZodBoolean>;
         }, z.core.$loose>>;
+        extension: z.ZodOptional<z.ZodObject<{
+            segmentSeconds: z.ZodNumber;
+            maxTotalSeconds: z.ZodNumber;
+        }, z.core.$loose>>;
     }, z.core.$loose>;
     defaults: z.ZodObject<{
         references: z.ZodOptional<z.ZodArray<z.ZodUnion<readonly [z.ZodObject<{

package/dist/catalog.d.ts

@@ -107,8 +107,14 @@ interface VideoRequest {
     watermark?: boolean;
     /** Provider-side safety checker toggle (fal MiniMax). Provider default when unset. */
     safetyChecker?: boolean;
-    /** Seedance omni-reference subtask hint. `edit` forces duration -1 and adaptive ratio. */
+    /** Seedance omni-reference subtask hint. `edit` forces duration -1 and adaptive ratio.
+     * `extend` on a `videoExtension` row continues the clip by `durationSeconds` more. */
     taskType?: "reference" | "edit" | "extend";
+    /** Continue a provider-side session instead of re-sending the clip (Gemini Omni
+     * Interactions `previous_interaction_id`). The prior call's `VideoResult.providerRef`
+     * carries it. The provider rejects it together with an explicit video task, so the
+     * adapter sends the prompt alone; `taskType: "extend"` still selects the extend path. */
+    previousInteractionId?: string;
     mode?: "standard" | "pro";
 }
 /**
@@ -160,7 +166,9 @@ interface MediaPart {
     cleanup?: () => Promise<void>;
 }
 type RequiredInput = "firstFrame" | "sourceVideo" | "references";
-type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "safetySettings" | "temperature" | "upscale" | "watermark" | "safetyChecker" | "taskType" | "instrumental";
+type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "generateAudio" | "promptExpansion" | "lastFrame" | "count" | "outputFormat" | "scaleFactor" | "safetySettings" | "temperature" | "upscale" | "watermark" | "safetyChecker" | "taskType" | "instrumental"
+/** Video rows that extend a clip in fixed segments up to a cumulative cap (see `ModelConstraints.extension`). */
+ | "videoExtension";
 interface VideoSecondsLimits {
     /** Minimum length of each reference video, in seconds. */
     min?: number;
@@ -210,6 +218,12 @@ interface ModelConstraints {
         /** The provider sends the source as a reference video, so it counts toward `references.videoSeconds.total` (Seedance). */
         inReferenceTotal?: boolean;
     };
+    /** Video extension (rows with `supports: ["videoExtension"]`): each `taskType: "extend"`
+     * call adds up to `segmentSeconds` and the clip can grow to `maxTotalSeconds`. */
+    extension?: {
+        segmentSeconds: number;
+        maxTotalSeconds: number;
+    };
 }
 /** JSON-compatible value. Provider options and raw payloads carry these, never `unknown`. */
 type JsonPrimitive = string | number | boolean | null;
@@ -351,6 +365,10 @@ interface VideoResult {
     };
     usage: UsageEvent;
     raw?: unknown;
+    /** Provider handle for chaining a follow-up call (`VideoRequest.previousInteractionId`). */
+    providerRef?: {
+        interactionId: string;
+    };
 }
 interface VideoCallJob extends VideoJob {
     warnings: Warning[];
@@ -613,6 +631,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             safetyChecker: "safetyChecker";
             taskType: "taskType";
             instrumental: "instrumental";
+            videoExtension: "videoExtension";
         }>>>;
         ratioIgnoredWith: z.ZodOptional<z.ZodArray<z.ZodEnum<{
             firstFrame: "firstFrame";
@@ -624,6 +643,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             editMin: z.ZodOptional<z.ZodNumber>;
             inReferenceTotal: z.ZodOptional<z.ZodBoolean>;
         }, z.core.$loose>>;
+        extension: z.ZodOptional<z.ZodObject<{
+            segmentSeconds: z.ZodNumber;
+            maxTotalSeconds: z.ZodNumber;
+        }, z.core.$loose>>;
     }, z.core.$loose>;
     defaults: z.ZodObject<{
         references: z.ZodOptional<z.ZodArray<z.ZodUnion<readonly [z.ZodObject<{

package/dist/catalog.js

@@ -1391,7 +1391,14 @@ 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 },
@@ -1399,7 +1406,7 @@ 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 }
   },
@@ -2130,6 +2137,7 @@ var SupportedOptionSchema = z2.enum([
   "watermark",
   "safetyChecker",
   "taskType",
+  "videoExtension",
   "instrumental"
 ]);
 var UsageUnitSchema = z2.enum([
@@ -2177,7 +2185,8 @@ var ModelConstraintsSchema = z2.looseObject({
   requires: z2.array(RequiredInputSchema).optional(),
   supports: z2.array(SupportedOptionSchema).optional(),
   ratioIgnoredWith: z2.array(z2.enum(["firstFrame", "sourceVideo"])).optional(),
-  sourceVideoSeconds: VideoSecondsLimitsSchema.omit({ total: true }).extend({ inReferenceTotal: z2.boolean().optional() }).optional()
+  sourceVideoSeconds: VideoSecondsLimitsSchema.omit({ total: true }).extend({ inReferenceTotal: z2.boolean().optional() }).optional(),
+  extension: z2.looseObject({ segmentSeconds: z2.number(), maxTotalSeconds: z2.number() }).optional()
 });
 var MediaRefSchema = z2.union([
   z2.looseObject({
@@ -2924,6 +2933,7 @@ 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",
@@ -2933,8 +2943,17 @@ 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") {