@shop-os/modelkit 0.2.25 → 0.2.26

package/dist/index.d.cts

@@ -890,8 +890,9 @@ declare const CatalogModelSchema: z.ZodType<CatalogModel>;
 declare const CatalogSchema: z.ZodType<CatalogModel[]>;
 
 declare const VIDEO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "aspectRatio", "resolution", "firstFrame", "lastFrame", "seed", "negativePrompt", "generateAudio", "cfgScale", "promptExpansion", "referenceImages"];
+declare const IMAGE_INPUT_SPEC_KEYS: readonly ["prompt", "references", "aspectRatio", "size", "outputFormat", "seed"];
 declare const AUDIO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "seed", "negativePrompt", "references", "sourceVideo", "instrumental"];
-type InputSpecKey = (typeof VIDEO_INPUT_SPEC_KEYS)[number] | (typeof AUDIO_INPUT_SPEC_KEYS)[number];
+type InputSpecKey = (typeof VIDEO_INPUT_SPEC_KEYS)[number] | (typeof IMAGE_INPUT_SPEC_KEYS)[number] | (typeof AUDIO_INPUT_SPEC_KEYS)[number];
 declare const INPUT_SPEC_SUPPORTS_KEYS: readonly SupportedOption[];
 declare const InputSpecSchema: z.ZodObject<{
     fields: z.ZodRecord<z.ZodString, z.ZodString>;
@@ -1243,4 +1244,4 @@ 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 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 KeyTier, 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, missingTierKeys, 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, 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 InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, 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, missingTierKeys, 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

@@ -890,8 +890,9 @@ declare const CatalogModelSchema: z.ZodType<CatalogModel>;
 declare const CatalogSchema: z.ZodType<CatalogModel[]>;
 
 declare const VIDEO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "aspectRatio", "resolution", "firstFrame", "lastFrame", "seed", "negativePrompt", "generateAudio", "cfgScale", "promptExpansion", "referenceImages"];
+declare const IMAGE_INPUT_SPEC_KEYS: readonly ["prompt", "references", "aspectRatio", "size", "outputFormat", "seed"];
 declare const AUDIO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "seed", "negativePrompt", "references", "sourceVideo", "instrumental"];
-type InputSpecKey = (typeof VIDEO_INPUT_SPEC_KEYS)[number] | (typeof AUDIO_INPUT_SPEC_KEYS)[number];
+type InputSpecKey = (typeof VIDEO_INPUT_SPEC_KEYS)[number] | (typeof IMAGE_INPUT_SPEC_KEYS)[number] | (typeof AUDIO_INPUT_SPEC_KEYS)[number];
 declare const INPUT_SPEC_SUPPORTS_KEYS: readonly SupportedOption[];
 declare const InputSpecSchema: z.ZodObject<{
     fields: z.ZodRecord<z.ZodString, z.ZodString>;
@@ -1243,4 +1244,4 @@ 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 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 KeyTier, 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, missingTierKeys, 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, 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 InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, 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, missingTierKeys, 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.js

@@ -2093,6 +2093,14 @@ var VIDEO_INPUT_SPEC_KEYS = [
   "promptExpansion",
   "referenceImages"
 ];
+var IMAGE_INPUT_SPEC_KEYS = [
+  "prompt",
+  "references",
+  "aspectRatio",
+  "size",
+  "outputFormat",
+  "seed"
+];
 var AUDIO_INPUT_SPEC_KEYS = [
   "prompt",
   "durationSeconds",
@@ -2109,14 +2117,22 @@ var INPUT_SPEC_SUPPORTS_KEYS = [
   "generateAudio",
   "cfgScale",
   "promptExpansion",
+  "outputFormat",
   "instrumental"
 ];
 var VIDEO_KEY_SET = new Set(VIDEO_INPUT_SPEC_KEYS);
+var IMAGE_KEY_SET = new Set(IMAGE_INPUT_SPEC_KEYS);
 var AUDIO_KEY_SET = new Set(AUDIO_INPUT_SPEC_KEYS);
 var ALL_KEY_SET = /* @__PURE__ */ new Set([
   ...VIDEO_INPUT_SPEC_KEYS,
+  ...IMAGE_INPUT_SPEC_KEYS,
   ...AUDIO_INPUT_SPEC_KEYS
 ]);
+var ALLOWED_KEYS_BY_TYPE = /* @__PURE__ */ new Map([
+  ["video", VIDEO_KEY_SET],
+  ["image", IMAGE_KEY_SET],
+  ["audio", AUDIO_KEY_SET]
+]);
 var ProviderScalarSchema = z2.union([z2.string(), z2.number(), z2.boolean()]);
 var FieldsSchema = z2.record(z2.string(), z2.string().min(1)).superRefine((fields, ctx) => {
   const keys = Object.keys(fields);
@@ -2185,13 +2201,13 @@ function validateInputSpec(raw, row) {
     };
   }
   const issues = [];
-  if (row.type !== "video" && row.type !== "audio") {
+  const allowed = ALLOWED_KEYS_BY_TYPE.get(row.type);
+  if (allowed === void 0) {
     issues.push({
       path: [],
       message: `inputSpec is not supported for "${row.type}" rows`
     });
   } else {
-    const allowed = row.type === "video" ? VIDEO_KEY_SET : AUDIO_KEY_SET;
     for (const key of Object.keys(spec.fields)) {
       if (!allowed.has(key)) {
         issues.push({
@@ -6735,6 +6751,12 @@ async function generateReplicateImage(req, ctx, deps) {
       break;
     }
     default: {
+      const specResult = replicateInputSpec(req, ctx);
+      if (specResult.kind === "invalid") throw replicateInvalidSpec(ctx, specResult.issues);
+      if (specResult.kind === "ok") {
+        input = specResult.input;
+        break;
+      }
       const first = (req.references ?? [])[0];
       const imageInput = { prompt: req.prompt };
       if (first) imageInput.image = refToUrl(first);
@@ -8815,6 +8837,7 @@ export {
   CatalogSchema,
   DEFAULT_ADAPTERS,
   ExposureSchema,
+  IMAGE_INPUT_SPEC_KEYS,
   INPUT_SPEC_SUPPORTS_KEYS,
   InputSpecSchema,
   JsonObjectSchema,