@shop-os/modelkit 0.13.0 → 0.15.0

package/dist/index.d.cts

@@ -132,7 +132,7 @@ interface Model3dConstraints {
     };
 }
 /** What a `video-tool` row does to its source clip. */
-type VideoToolKind = "upscale" | "interpolate" | "hdr" | "bg-remove" | "subtitles";
+type VideoToolKind = "upscale" | "interpolate" | "hdr" | "bg-remove" | "subtitles" | "depth";
 /** Container/codec a video-tool row can write. */
 /** `exr`: a ZIP of 16-bit EXR frames in `VideoToolResult.frames`, with an MP4 preview as `video`. */
 type VideoToolOutputFormat = "h265" | "prores" | "exr";
@@ -601,6 +601,11 @@ type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "gen
  | "insertWindow"
 /** Mask rows: `ImageRequest.points` and `ImageRequest.boxes`. */
  | "points" | "boxes";
+/** Smallest frame a provider takes for an input video, in pixels (fal: "Min width: 700px"). */
+interface VideoPixelLimits {
+    minWidth?: number;
+    minHeight?: number;
+}
 interface VideoSecondsLimits {
     /** Minimum length of each reference video, in seconds. */
     min?: number;
@@ -622,6 +627,8 @@ interface ReferenceLimits {
     audioOnly?: boolean;
     /** Per-video length caps (min/max each) and combined total for reference videos. */
     videoSeconds?: VideoSecondsLimits;
+    /** Smallest frame each reference video may have (Kling 4.0: 700 px). Kits before 0.15.0 ignore it. */
+    videoPixels?: VideoPixelLimits;
     /** Per-clip length caps (min/max each) and combined total for reference audio, in seconds. */
     audioSeconds?: {
         min?: number;
@@ -751,6 +758,8 @@ interface ModelConstraints {
     languages?: LanguageOption[];
     /** Audio-output rows whose provider lets the caller pick a file format; `defaults.outputFormat` is one. */
     outputFormats?: string[];
+    /** Smallest frame the source video may have (Kling 4.0 Edit: 700 px). Kits before 0.15.0 ignore it. */
+    sourceVideoPixels?: VideoPixelLimits;
     /** Length limits for the source video the row edits or extends. */
     sourceVideoSeconds?: Omit<VideoSecondsLimits, "total"> & {
         /** The provider sends the source as a reference video, so it counts toward `references.videoSeconds.total` (Seedance). */
@@ -1253,6 +1262,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
                 total: z.ZodOptional<z.ZodNumber>;
                 editMin: z.ZodOptional<z.ZodNumber>;
             }, z.core.$loose>>;
+            videoPixels: z.ZodOptional<z.ZodObject<{
+                minWidth: z.ZodOptional<z.ZodNumber>;
+                minHeight: z.ZodOptional<z.ZodNumber>;
+            }, z.core.$loose>>;
             audioSeconds: z.ZodOptional<z.ZodObject<{
                 min: z.ZodOptional<z.ZodNumber>;
                 max: z.ZodOptional<z.ZodNumber>;
@@ -1330,6 +1343,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             name: z.ZodString;
         }, z.core.$loose>>>;
         outputFormats: z.ZodOptional<z.ZodArray<z.ZodString>>;
+        sourceVideoPixels: z.ZodOptional<z.ZodObject<{
+            minWidth: z.ZodOptional<z.ZodNumber>;
+            minHeight: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$loose>>;
         sourceVideoSeconds: z.ZodOptional<z.ZodObject<{
             max: z.ZodOptional<z.ZodNumber>;
             min: z.ZodOptional<z.ZodNumber>;
@@ -1351,6 +1368,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
                 interpolate: "interpolate";
                 hdr: "hdr";
                 subtitles: "subtitles";
+                depth: "depth";
             }>;
             upscaleFactor: z.ZodOptional<z.ZodObject<{
                 min: z.ZodNumber;
@@ -1857,6 +1875,19 @@ declare function videoLengthViolation(model: CatalogModel, input: {
     sourceVideo?: number;
     taskType?: VideoRequest["taskType"];
 }): string | undefined;
+type MeasuredFrame = {
+    width?: number;
+    height?: number;
+} | undefined;
+/**
+ * First frame-size violation for the videos a request sends, as a user-facing message, from
+ * their measured pixel sizes (`references.videoPixels`, `sourceVideoPixels`). An unmeasured
+ * video passes: the provider still has the last word.
+ */
+declare function videoSizeViolation(model: CatalogModel, input: {
+    referenceVideos?: MeasuredFrame[];
+    sourceVideo?: MeasuredFrame;
+}): string | undefined;
 declare function resolveImageOptions(model: CatalogModel, request: ImageRequest, onUnsupported?: OnUnsupported): ResolveResult<ImageRequest>;
 declare function resolveVideoOptions(model: CatalogModel, request: VideoRequest, onUnsupported?: OnUnsupported): ResolveResult<VideoRequest>;
 declare function resolveAudioOptions(model: CatalogModel, request: AudioRequest, onUnsupported?: OnUnsupported): ResolveResult<AudioRequest>;
@@ -2262,4 +2293,4 @@ declare function describeThrown(cause: unknown): ThrownErrorInfo;
 /** Add the AI SDK `GeneratedFile` accessors to adapter image bytes. base64 is computed on first read. */
 declare function toGeneratedImage(image: ImageBytes): GeneratedImage;
 
-export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };
+export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoPixelLimits, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoSizeViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.d.ts

@@ -132,7 +132,7 @@ interface Model3dConstraints {
     };
 }
 /** What a `video-tool` row does to its source clip. */
-type VideoToolKind = "upscale" | "interpolate" | "hdr" | "bg-remove" | "subtitles";
+type VideoToolKind = "upscale" | "interpolate" | "hdr" | "bg-remove" | "subtitles" | "depth";
 /** Container/codec a video-tool row can write. */
 /** `exr`: a ZIP of 16-bit EXR frames in `VideoToolResult.frames`, with an MP4 preview as `video`. */
 type VideoToolOutputFormat = "h265" | "prores" | "exr";
@@ -601,6 +601,11 @@ type SupportedOption = "seed" | "quality" | "negativePrompt" | "cfgScale" | "gen
  | "insertWindow"
 /** Mask rows: `ImageRequest.points` and `ImageRequest.boxes`. */
  | "points" | "boxes";
+/** Smallest frame a provider takes for an input video, in pixels (fal: "Min width: 700px"). */
+interface VideoPixelLimits {
+    minWidth?: number;
+    minHeight?: number;
+}
 interface VideoSecondsLimits {
     /** Minimum length of each reference video, in seconds. */
     min?: number;
@@ -622,6 +627,8 @@ interface ReferenceLimits {
     audioOnly?: boolean;
     /** Per-video length caps (min/max each) and combined total for reference videos. */
     videoSeconds?: VideoSecondsLimits;
+    /** Smallest frame each reference video may have (Kling 4.0: 700 px). Kits before 0.15.0 ignore it. */
+    videoPixels?: VideoPixelLimits;
     /** Per-clip length caps (min/max each) and combined total for reference audio, in seconds. */
     audioSeconds?: {
         min?: number;
@@ -751,6 +758,8 @@ interface ModelConstraints {
     languages?: LanguageOption[];
     /** Audio-output rows whose provider lets the caller pick a file format; `defaults.outputFormat` is one. */
     outputFormats?: string[];
+    /** Smallest frame the source video may have (Kling 4.0 Edit: 700 px). Kits before 0.15.0 ignore it. */
+    sourceVideoPixels?: VideoPixelLimits;
     /** Length limits for the source video the row edits or extends. */
     sourceVideoSeconds?: Omit<VideoSecondsLimits, "total"> & {
         /** The provider sends the source as a reference video, so it counts toward `references.videoSeconds.total` (Seedance). */
@@ -1253,6 +1262,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
                 total: z.ZodOptional<z.ZodNumber>;
                 editMin: z.ZodOptional<z.ZodNumber>;
             }, z.core.$loose>>;
+            videoPixels: z.ZodOptional<z.ZodObject<{
+                minWidth: z.ZodOptional<z.ZodNumber>;
+                minHeight: z.ZodOptional<z.ZodNumber>;
+            }, z.core.$loose>>;
             audioSeconds: z.ZodOptional<z.ZodObject<{
                 min: z.ZodOptional<z.ZodNumber>;
                 max: z.ZodOptional<z.ZodNumber>;
@@ -1330,6 +1343,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             name: z.ZodString;
         }, z.core.$loose>>>;
         outputFormats: z.ZodOptional<z.ZodArray<z.ZodString>>;
+        sourceVideoPixels: z.ZodOptional<z.ZodObject<{
+            minWidth: z.ZodOptional<z.ZodNumber>;
+            minHeight: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$loose>>;
         sourceVideoSeconds: z.ZodOptional<z.ZodObject<{
             max: z.ZodOptional<z.ZodNumber>;
             min: z.ZodOptional<z.ZodNumber>;
@@ -1351,6 +1368,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
                 interpolate: "interpolate";
                 hdr: "hdr";
                 subtitles: "subtitles";
+                depth: "depth";
             }>;
             upscaleFactor: z.ZodOptional<z.ZodObject<{
                 min: z.ZodNumber;
@@ -1857,6 +1875,19 @@ declare function videoLengthViolation(model: CatalogModel, input: {
     sourceVideo?: number;
     taskType?: VideoRequest["taskType"];
 }): string | undefined;
+type MeasuredFrame = {
+    width?: number;
+    height?: number;
+} | undefined;
+/**
+ * First frame-size violation for the videos a request sends, as a user-facing message, from
+ * their measured pixel sizes (`references.videoPixels`, `sourceVideoPixels`). An unmeasured
+ * video passes: the provider still has the last word.
+ */
+declare function videoSizeViolation(model: CatalogModel, input: {
+    referenceVideos?: MeasuredFrame[];
+    sourceVideo?: MeasuredFrame;
+}): string | undefined;
 declare function resolveImageOptions(model: CatalogModel, request: ImageRequest, onUnsupported?: OnUnsupported): ResolveResult<ImageRequest>;
 declare function resolveVideoOptions(model: CatalogModel, request: VideoRequest, onUnsupported?: OnUnsupported): ResolveResult<VideoRequest>;
 declare function resolveAudioOptions(model: CatalogModel, request: AudioRequest, onUnsupported?: OnUnsupported): ResolveResult<AudioRequest>;
@@ -2262,4 +2293,4 @@ declare function describeThrown(cause: unknown): ThrownErrorInfo;
 /** Add the AI SDK `GeneratedFile` accessors to adapter image bytes. base64 is computed on first read. */
 declare function toGeneratedImage(image: ImageBytes): GeneratedImage;
 
-export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };
+export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoPixelLimits, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoSizeViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.js

@@ -5093,6 +5093,7 @@ function klingMotionControl(tier) {
         videos: 1,
         audios: 0,
         videoSeconds: { min: 3, max: 30.05 },
+        videoPixels: { minWidth: 340, minHeight: 340 },
         min: { images: 1, videos: 1 }
       },
       requires: ["references"],
@@ -5159,6 +5160,7 @@ var VIDEO_FAMILY_MODELS = [
       references: { ...NO_REFERENCES, images: 4 },
       requires: ["sourceVideo"],
       sourceVideoSeconds: { min: 3, max: 15.05 },
+      sourceVideoPixels: { minWidth: 720, minHeight: 720 },
       // generateAudio keeps (true) or drops (false) the source soundtrack.
       supports: ["generateAudio"],
       prompt: KLING_PROMPT,
@@ -5199,7 +5201,8 @@ var VIDEO_FAMILY_MODELS = [
         ...NO_REFERENCES,
         images: 10,
         videos: 5,
-        videoSeconds: { min: 3, max: 30.05, total: 30 }
+        videoSeconds: { min: 3, max: 30.05, total: 30 },
+        videoPixels: { minWidth: 700, minHeight: 700 }
       },
       supports: ["lastFrame"],
       prompt: KLING_V4_PROMPT,
@@ -5233,6 +5236,7 @@ var VIDEO_FAMILY_MODELS = [
       references: { ...NO_REFERENCES, images: 10 },
       requires: ["sourceVideo"],
       sourceVideoSeconds: { min: 3, max: 30.05 },
+      sourceVideoPixels: { minWidth: 700, minHeight: 700 },
       supports: [],
       prompt: KLING_V4_PROMPT,
       audio: KLING_V4_AUDIO
@@ -6569,6 +6573,33 @@ var VIDEO_TOOL_MODELS = [
       description: "Topaz Hyperion 2.5 SDR to HDR: a 10-bit HDR master in H.265 (HDR10) or ProRes 422 HQ, at the source's size and frame rate."
     }
   },
+  {
+    // https://fal.ai/models/fal-ai/depth-anything-video/api: greyscale depth (white = near)
+    // at the source's frame rate, short edge capped at 1080 px (a 2160x3840 source came
+    // back 1080x1920), at most 2400 frames; one fal price of $0.04 per second for every
+    // model size. The adapter always sends VDA-Small, the only Apache-2.0 size. 100 s is
+    // 2400 frames at 24 fps; the frame check refuses faster clips sooner.
+    id: "depth-anything-video",
+    aliases: [],
+    label: "Video Depth Anything",
+    type: "video-tool",
+    adapter: "fal",
+    providerModelId: "fal-ai/depth-anything-video",
+    lifecycle: "active",
+    exposure: ["cowork"],
+    constraints: {
+      supports: [],
+      sourceVideoSeconds: { max: 100 },
+      tool: { kind: "depth", maxOutputShortEdge: 1080, maxOutputFrames: 2400 }
+    },
+    defaults: {},
+    usage: SECONDS_USAGE,
+    pricingKind: "video_per_second",
+    display: {
+      provider: "depth-anything",
+      description: "Video Depth Anything: a greyscale depth map of the clip (near is white), frame for frame, for motion and camera control."
+    }
+  },
   {
     id: "flux-video-upscale-precise",
     aliases: [],
@@ -7750,6 +7781,10 @@ var VideoSecondsLimitsSchema = z3.looseObject({
   total: z3.number().optional(),
   editMin: z3.number().optional()
 });
+var VideoPixelLimitsSchema = z3.looseObject({
+  minWidth: z3.number().int().positive().optional(),
+  minHeight: z3.number().int().positive().optional()
+});
 var ReferenceLimitsSchema = z3.looseObject({
   images: z3.number().optional(),
   imagesWithVideo: z3.number().optional(),
@@ -7758,6 +7793,7 @@ var ReferenceLimitsSchema = z3.looseObject({
   total: z3.number().optional(),
   audioOnly: z3.boolean().optional(),
   videoSeconds: VideoSecondsLimitsSchema.optional(),
+  videoPixels: VideoPixelLimitsSchema.optional(),
   audioSeconds: z3.looseObject({
     min: z3.number().optional(),
     max: z3.number().optional(),
@@ -7786,7 +7822,7 @@ var LanguageOptionSchema = z3.looseObject({
   name: z3.string().min(1)
 });
 var VideoToolConstraintsSchema = z3.looseObject({
-  kind: z3.enum(["upscale", "interpolate", "hdr", "bg-remove", "subtitles"]),
+  kind: z3.enum(["upscale", "interpolate", "hdr", "bg-remove", "subtitles", "depth"]),
   upscaleFactor: RangeSchema.optional(),
   maxOutputLongEdge: z3.number().positive().optional(),
   prompt: z3.boolean().optional(),
@@ -7847,6 +7883,7 @@ var ModelConstraintsSchema = z3.looseObject({
   }).optional(),
   languages: z3.array(LanguageOptionSchema).optional(),
   outputFormats: z3.array(z3.string().min(1)).min(1).optional(),
+  sourceVideoPixels: VideoPixelLimitsSchema.optional(),
   sourceVideoSeconds: VideoSecondsLimitsSchema.omit({ total: true }).extend({ inReferenceTotal: z3.boolean().optional() }).optional(),
   // segmentSeconds/maxTotalSeconds stay required: a kit before 0.3.0 drops a row whose
   // extension lacks them, so every extension block keeps them.
@@ -9556,7 +9593,7 @@ function checkedQueueUrl(ctx, url, what) {
 import { z as z12 } from "zod";
 var RangeSchema2 = z12.object({ min: z12.number(), max: z12.number() });
 var ToolSchema = z12.object({
-  kind: z12.enum(["upscale", "interpolate", "hdr", "bg-remove", "subtitles"]),
+  kind: z12.enum(["upscale", "interpolate", "hdr", "bg-remove", "subtitles", "depth"]),
   upscaleFactor: RangeSchema2.optional(),
   maxOutputLongEdge: z12.number().positive().optional(),
   prompt: z12.boolean().optional(),
@@ -9694,6 +9731,25 @@ function planVideoTool(row2, req) {
       }
     };
   }
+  if (tool.kind === "depth") {
+    const limit = tool.maxOutputFrames;
+    if (limit !== void 0) {
+      if (fps === void 0) return fail("The source video's frame rate is required");
+      if (durationSeconds * fps > limit) {
+        return fail(`${row2.label} renders up to ${limit} frames (${seconds(limit, fps)}).`);
+      }
+    }
+    const out2 = shrinkToShortEdge(width, height, tool.maxOutputShortEdge);
+    return {
+      ok: true,
+      output: {
+        ...base,
+        ...out2,
+        resolution: resolutionTier(out2.width, out2.height),
+        format: "h264"
+      }
+    };
+  }
   if (tool.kind === "subtitles") {
     const resolution = resolutionTier(width, height);
     const problem = subtitleProblem(
@@ -9790,12 +9846,15 @@ function toolInput(ctx, req, output) {
     input["output_codec"] = "vp9";
   } else if (tool?.kind === "subtitles") {
     Object.assign(input, subtitleFields(tool, req.subtitles));
+  } else if (tool?.kind === "depth") {
+    input["model"] = "VDA-Small";
+    input["colormap"] = "grayscale";
   } else if (output.format === "exr") {
     const prompt = tool?.prompt === true ? req.prompt?.trim() : void 0;
     input["prompt"] = prompt || EXR_DEFAULT_PROMPT;
     if (output.fps !== void 0) input["frames_per_second"] = output.fps;
     if (output.frames !== void 0) input["num_frames"] = output.frames;
-  } else {
+  } else if (tool?.kind === "hdr") {
     input["output_format"] = output.format === "prores" ? "prores" : "mp4";
   }
   return input;
@@ -17567,6 +17626,28 @@ function videoLengthViolation(model, input) {
     return `${model.label} takes a source video of at least ${String(sourceMin)} s; this one is ${seconds2(source2, sourceMin)} s.`;
   return void 0;
 }
+function undersizedSide(frame, limits) {
+  if (!frame || !limits) return void 0;
+  const { width, height } = frame;
+  if (limits.minWidth !== void 0 && width !== void 0 && width > 0 && width < limits.minWidth)
+    return { side: "wide", min: limits.minWidth, actual: width };
+  if (limits.minHeight !== void 0 && height !== void 0 && height > 0 && height < limits.minHeight)
+    return { side: "tall", min: limits.minHeight, actual: height };
+  return void 0;
+}
+function videoSizeViolation(model, input) {
+  const refLimits = model.constraints.references?.videoPixels;
+  const refs = input.referenceVideos ?? [];
+  for (let i = 0; i < refs.length; i++) {
+    const under2 = undersizedSide(refs[i], refLimits);
+    if (under2)
+      return `${model.label} takes reference videos at least ${String(under2.min)} px ${under2.side}; video ${i + 1} is ${String(under2.actual)} px.`;
+  }
+  const under = undersizedSide(input.sourceVideo, model.constraints.sourceVideoPixels);
+  if (under)
+    return `${model.label} takes a source video at least ${String(under.min)} px ${under.side}; this one is ${String(under.actual)} px.`;
+  return void 0;
+}
 function resolveImageOptions(model, request, onUnsupported = "error") {
   const warnings = [];
   const c = model.constraints;
@@ -19686,6 +19767,7 @@ export {
   toGeneratedImage,
   validateInputSpec,
   videoLengthViolation,
+  videoSizeViolation,
   videoToolConstraints,
   withRetry,
   xaiVideoAdapter