@shop-os/modelkit 0.2.19 → 0.2.21

package/dist/index.d.cts

@@ -194,6 +194,15 @@ interface ReferenceLimits {
     /** Per-video length caps (min/max each) and combined total for reference videos. */
     videoSeconds?: VideoSecondsLimits;
 }
+/** Whether a video row's output has sound: always, switchable (`supports: generateAudio`), or silent. */
+type AudioOutput = "always" | "optional" | "never";
+/** How a prompt points at reference N: "image N" numbering, a caption beside each image, or prose. */
+type ReferenceSyntax = "numbered" | "caption" | "descriptive";
+interface PromptConventions {
+    referenceSyntax: ReferenceSyntax;
+    /** Provider-documented prompt length limit, in characters. */
+    maxChars?: number;
+}
 interface ModelConstraints {
     aspectRatios?: AspectRatio[];
     sizes?: ImageSize[];
@@ -227,6 +236,10 @@ interface ModelConstraints {
         segmentSeconds: number;
         maxTotalSeconds: number;
     };
+    /** Whether the output has sound. Read it through `outputAudio(row)`, which tolerates bad data. */
+    audio?: AudioOutput;
+    /** Prompt conventions. Read them through `promptConventions(row)`, which fills the default. */
+    prompt?: PromptConventions;
 }
 /** JSON-compatible value. Provider options and raw payloads carry these, never `unknown`. */
 type JsonPrimitive = string | number | boolean | null;
@@ -285,6 +298,8 @@ interface CatalogCapabilities {
     acceptsSourceVideo?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
+    /** Video rows: a last frame is accepted only together with a first frame. */
+    lastFrameRequiresFirstFrame?: boolean;
     /** Text rows. `reasoning`: the model streams reasoning the UI should surface. */
     reasoning?: boolean;
     vision?: boolean;
@@ -674,6 +689,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             segmentSeconds: z.ZodNumber;
             maxTotalSeconds: z.ZodNumber;
         }, z.core.$loose>>;
+        audio: z.ZodOptional<z.ZodUnknown>;
+        prompt: z.ZodOptional<z.ZodUnknown>;
     }, z.core.$loose>;
     defaults: z.ZodObject<{
         references: z.ZodOptional<z.ZodArray<z.ZodUnion<readonly [z.ZodObject<{
@@ -860,6 +877,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
     capabilities: z.ZodOptional<z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
+        lastFrameRequiresFirstFrame: z.ZodOptional<z.ZodBoolean>;
         reasoning: z.ZodOptional<z.ZodBoolean>;
         vision: z.ZodOptional<z.ZodBoolean>;
         contextWindow: z.ZodOptional<z.ZodNumber>;
@@ -1219,4 +1237,4 @@ 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, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type AspectRatio, 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 Exposure, ExposureSchema, type GeneratedImage, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type LanguageModelHandle, type Lifecycle, type LiveCatalogState, MODELS, type MediaPart, type MediaRef, 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 ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, type VideoResult, type VideoSecondsLimits, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, withRetry, xaiVideoAdapter };
+export { ADAPTER_IDS, ADAPTER_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type AspectRatio, 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 Exposure, ExposureSchema, type GeneratedImage, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type LanguageModelHandle, type Lifecycle, type LiveCatalogState, MODELS, type MediaPart, type MediaRef, 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 SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, type VideoResult, type VideoSecondsLimits, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, withRetry, xaiVideoAdapter };

package/dist/index.d.ts

@@ -194,6 +194,15 @@ interface ReferenceLimits {
     /** Per-video length caps (min/max each) and combined total for reference videos. */
     videoSeconds?: VideoSecondsLimits;
 }
+/** Whether a video row's output has sound: always, switchable (`supports: generateAudio`), or silent. */
+type AudioOutput = "always" | "optional" | "never";
+/** How a prompt points at reference N: "image N" numbering, a caption beside each image, or prose. */
+type ReferenceSyntax = "numbered" | "caption" | "descriptive";
+interface PromptConventions {
+    referenceSyntax: ReferenceSyntax;
+    /** Provider-documented prompt length limit, in characters. */
+    maxChars?: number;
+}
 interface ModelConstraints {
     aspectRatios?: AspectRatio[];
     sizes?: ImageSize[];
@@ -227,6 +236,10 @@ interface ModelConstraints {
         segmentSeconds: number;
         maxTotalSeconds: number;
     };
+    /** Whether the output has sound. Read it through `outputAudio(row)`, which tolerates bad data. */
+    audio?: AudioOutput;
+    /** Prompt conventions. Read them through `promptConventions(row)`, which fills the default. */
+    prompt?: PromptConventions;
 }
 /** JSON-compatible value. Provider options and raw payloads carry these, never `unknown`. */
 type JsonPrimitive = string | number | boolean | null;
@@ -285,6 +298,8 @@ interface CatalogCapabilities {
     acceptsSourceVideo?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
+    /** Video rows: a last frame is accepted only together with a first frame. */
+    lastFrameRequiresFirstFrame?: boolean;
     /** Text rows. `reasoning`: the model streams reasoning the UI should surface. */
     reasoning?: boolean;
     vision?: boolean;
@@ -674,6 +689,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             segmentSeconds: z.ZodNumber;
             maxTotalSeconds: z.ZodNumber;
         }, z.core.$loose>>;
+        audio: z.ZodOptional<z.ZodUnknown>;
+        prompt: z.ZodOptional<z.ZodUnknown>;
     }, z.core.$loose>;
     defaults: z.ZodObject<{
         references: z.ZodOptional<z.ZodArray<z.ZodUnion<readonly [z.ZodObject<{
@@ -860,6 +877,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
     capabilities: z.ZodOptional<z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
+        lastFrameRequiresFirstFrame: z.ZodOptional<z.ZodBoolean>;
         reasoning: z.ZodOptional<z.ZodBoolean>;
         vision: z.ZodOptional<z.ZodBoolean>;
         contextWindow: z.ZodOptional<z.ZodNumber>;
@@ -1219,4 +1237,4 @@ 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, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type AspectRatio, 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 Exposure, ExposureSchema, type GeneratedImage, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type LanguageModelHandle, type Lifecycle, type LiveCatalogState, MODELS, type MediaPart, type MediaRef, 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 ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, type VideoResult, type VideoSecondsLimits, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, withRetry, xaiVideoAdapter };
+export { ADAPTER_IDS, ADAPTER_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type AspectRatio, 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 Exposure, ExposureSchema, type GeneratedImage, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type LanguageModelHandle, type Lifecycle, type LiveCatalogState, MODELS, type MediaPart, type MediaRef, 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 SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, type VideoResult, type VideoSecondsLimits, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, withRetry, xaiVideoAdapter };