@shop-os/modelkit 0.3.1 → 0.3.3

package/dist/index.d.cts

@@ -11,7 +11,7 @@ import { LanguageModelV3, LanguageModelV2 } from '@ai-sdk/provider';
 
 /** The AI SDK video model spelling: a catalog id or a provider video model instance. */
 type VideoModel = Parameters<typeof experimental_generateVideo>[0]["model"];
-type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "adaptive";
+type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "1:4" | "4:1" | "1:8" | "8:1" | "adaptive";
 type ImageSize = "512" | "1K" | "2K" | "4K";
 type VideoResolution = "360p" | "480p" | "720p" | "768p" | "1080p" | "2k" | "4k";
 /** "xhigh" and "max" exist only on gpt-image-2.5; gpt-image-2 rejects them at the provider. */
@@ -396,6 +396,8 @@ interface CatalogCapabilities {
     videoEdit?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
+    /** Video rows: a source video is never sent with first/last frames or reference images. */
+    sourceVideoExcludesFrames?: 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. */
@@ -716,6 +718,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             "21:9": "21:9";
             "9:21": "9:21";
             "97:30": "97:30";
+            "1:4": "1:4";
+            "4:1": "4:1";
+            "1:8": "1:8";
+            "8:1": "8:1";
             adaptive: "adaptive";
         }>>>;
         sizes: z.ZodOptional<z.ZodArray<z.ZodEnum<{
@@ -851,6 +857,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             "21:9": "21:9";
             "9:21": "9:21";
             "97:30": "97:30";
+            "1:4": "1:4";
+            "4:1": "4:1";
+            "1:8": "1:8";
+            "8:1": "8:1";
             adaptive: "adaptive";
         }>>;
         size: z.ZodOptional<z.ZodEnum<{
@@ -1013,6 +1023,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
         videoEdit: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
+        sourceVideoExcludesFrames: z.ZodOptional<z.ZodBoolean>;
         lastFrameRequiresFirstFrame: z.ZodOptional<z.ZodBoolean>;
         reasoning: z.ZodOptional<z.ZodBoolean>;
         vision: z.ZodOptional<z.ZodBoolean>;
@@ -1022,15 +1033,19 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
 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 VIDEO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "aspectRatio", "resolution", "firstFrame", "lastFrame", "seed", "negativePrompt", "generateAudio", "cfgScale", "promptExpansion", "referenceImages", "sourceVideo"];
 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"];
+/** Image inputs that `inputSpec.arrays` can join into one provider array field. */
+declare const ARRAY_INPUT_SPEC_KEYS: readonly ["firstFrame", "referenceImages", "lastFrame", "references"];
+type ArrayInputSpecKey = (typeof ARRAY_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>;
     values: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnion<readonly [z.ZodString, z.ZodNumber, z.ZodBoolean]>>>>;
     static: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnion<readonly [z.ZodString, z.ZodNumber, z.ZodBoolean]>>>;
+    arrays: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodArray<z.ZodString>>>;
     editEndpoint: z.ZodOptional<z.ZodString>;
 }, z.core.$strict>;
 type InputSpec = z.infer<typeof InputSpecSchema>;
@@ -1046,6 +1061,7 @@ interface ValidateRow {
     };
 }
 declare function isInputSpecKey(key: string): key is InputSpecKey;
+declare function isArrayInputSpecKey(key: string): key is ArrayInputSpecKey;
 declare function validateInputSpec(raw: ProviderOptionValue, row: ValidateRow): {
     ok: true;
     spec: InputSpec;
@@ -1437,4 +1453,4 @@ 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 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 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, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, 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 Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, missingTierKeys, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, 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 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 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 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, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, 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 Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, missingTierKeys, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.d.ts

@@ -11,7 +11,7 @@ import { LanguageModelV3, LanguageModelV2 } from '@ai-sdk/provider';
 
 /** The AI SDK video model spelling: a catalog id or a provider video model instance. */
 type VideoModel = Parameters<typeof experimental_generateVideo>[0]["model"];
-type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "adaptive";
+type AspectRatio = "1:1" | "16:9" | "9:16" | "4:3" | "3:4" | "3:2" | "2:3" | "4:5" | "5:4" | "21:9" | "9:21" | "97:30" | "1:4" | "4:1" | "1:8" | "8:1" | "adaptive";
 type ImageSize = "512" | "1K" | "2K" | "4K";
 type VideoResolution = "360p" | "480p" | "720p" | "768p" | "1080p" | "2k" | "4k";
 /** "xhigh" and "max" exist only on gpt-image-2.5; gpt-image-2 rejects them at the provider. */
@@ -396,6 +396,8 @@ interface CatalogCapabilities {
     videoEdit?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
+    /** Video rows: a source video is never sent with first/last frames or reference images. */
+    sourceVideoExcludesFrames?: 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. */
@@ -716,6 +718,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             "21:9": "21:9";
             "9:21": "9:21";
             "97:30": "97:30";
+            "1:4": "1:4";
+            "4:1": "4:1";
+            "1:8": "1:8";
+            "8:1": "8:1";
             adaptive: "adaptive";
         }>>>;
         sizes: z.ZodOptional<z.ZodArray<z.ZodEnum<{
@@ -851,6 +857,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             "21:9": "21:9";
             "9:21": "9:21";
             "97:30": "97:30";
+            "1:4": "1:4";
+            "4:1": "4:1";
+            "1:8": "1:8";
+            "8:1": "8:1";
             adaptive: "adaptive";
         }>>;
         size: z.ZodOptional<z.ZodEnum<{
@@ -1013,6 +1023,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
         videoEdit: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
+        sourceVideoExcludesFrames: z.ZodOptional<z.ZodBoolean>;
         lastFrameRequiresFirstFrame: z.ZodOptional<z.ZodBoolean>;
         reasoning: z.ZodOptional<z.ZodBoolean>;
         vision: z.ZodOptional<z.ZodBoolean>;
@@ -1022,15 +1033,19 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
 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 VIDEO_INPUT_SPEC_KEYS: readonly ["prompt", "durationSeconds", "aspectRatio", "resolution", "firstFrame", "lastFrame", "seed", "negativePrompt", "generateAudio", "cfgScale", "promptExpansion", "referenceImages", "sourceVideo"];
 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"];
+/** Image inputs that `inputSpec.arrays` can join into one provider array field. */
+declare const ARRAY_INPUT_SPEC_KEYS: readonly ["firstFrame", "referenceImages", "lastFrame", "references"];
+type ArrayInputSpecKey = (typeof ARRAY_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>;
     values: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnion<readonly [z.ZodString, z.ZodNumber, z.ZodBoolean]>>>>;
     static: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnion<readonly [z.ZodString, z.ZodNumber, z.ZodBoolean]>>>;
+    arrays: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodArray<z.ZodString>>>;
     editEndpoint: z.ZodOptional<z.ZodString>;
 }, z.core.$strict>;
 type InputSpec = z.infer<typeof InputSpecSchema>;
@@ -1046,6 +1061,7 @@ interface ValidateRow {
     };
 }
 declare function isInputSpecKey(key: string): key is InputSpecKey;
+declare function isArrayInputSpecKey(key: string): key is ArrayInputSpecKey;
 declare function validateInputSpec(raw: ProviderOptionValue, row: ValidateRow): {
     ok: true;
     spec: InputSpec;
@@ -1437,4 +1453,4 @@ 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 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 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, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, 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 Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, missingTierKeys, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, 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 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 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 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, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, 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 Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, mergeTextProviderOptions, missingTierKeys, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveImageOptions, resolveKeys, resolveVideoOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.js

@@ -169,13 +169,21 @@ var GEMINI_RATIOS = [
   "5:4",
   "adaptive"
 ];
+var NANO_BANANA_2_1_RATIOS = [
+  ...GEMINI_RATIOS.filter((r) => r !== "adaptive"),
+  "1:4",
+  "4:1",
+  "1:8",
+  "8:1",
+  "adaptive"
+];
 var GEMINI_PROMPT_MAX_CHARS = 5e4;
 var GPT_IMAGE_PROMPT = { referenceSyntax: "numbered", maxChars: 32e3 };
-function geminiImage(id, label, aliases, sizes, display, exposure, references, promptMaxChars) {
+function geminiImage(id, label, aliases, sizes, display, exposure, references, promptMaxChars, aspectRatios = GEMINI_RATIOS) {
   const prompt = { referenceSyntax: "caption" };
   if (promptMaxChars !== void 0) prompt.maxChars = promptMaxChars;
   const constraints = {
-    aspectRatios: [...GEMINI_RATIOS],
+    aspectRatios: [...aspectRatios],
     sizes,
     supports: ["safetySettings", "temperature"],
     prompt
@@ -199,6 +207,21 @@ function geminiImage(id, label, aliases, sizes, display, exposure, references, p
   return model;
 }
 var IMAGE_MODELS = [
+  geminiImage(
+    "gemini-nano-banana-2.1",
+    "Nano Banana 2.1",
+    ["nb-2.1"],
+    ["1K", "2K", "4K"],
+    {
+      provider: "google",
+      description: "Latest high-efficiency Nano Banana. Better realism, prompt adherence, text rendering and character consistency at Flash speed. Great for campaigns, infographics, and iterative edits.",
+      logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png"
+    },
+    ["cowork", "space_builder"],
+    { images: 14 },
+    GEMINI_PROMPT_MAX_CHARS,
+    NANO_BANANA_2_1_RATIOS
+  ),
   geminiImage(
     "gemini-3-pro-image-preview",
     "Nano Banana Pro",
@@ -2215,7 +2238,8 @@ var VIDEO_INPUT_SPEC_KEYS = [
   "generateAudio",
   "cfgScale",
   "promptExpansion",
-  "referenceImages"
+  "referenceImages",
+  "sourceVideo"
 ];
 var IMAGE_INPUT_SPEC_KEYS = [
   "prompt",
@@ -2234,6 +2258,12 @@ var AUDIO_INPUT_SPEC_KEYS = [
   "sourceVideo",
   "instrumental"
 ];
+var ARRAY_INPUT_SPEC_KEYS = [
+  "firstFrame",
+  "referenceImages",
+  "lastFrame",
+  "references"
+];
 var INPUT_SPEC_SUPPORTS_KEYS = [
   "lastFrame",
   "seed",
@@ -2247,6 +2277,7 @@ var INPUT_SPEC_SUPPORTS_KEYS = [
 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 ARRAY_KEY_SET = new Set(ARRAY_INPUT_SPEC_KEYS);
 var ALL_KEY_SET = /* @__PURE__ */ new Set([
   ...VIDEO_INPUT_SPEC_KEYS,
   ...IMAGE_INPUT_SPEC_KEYS,
@@ -2291,13 +2322,30 @@ var ValuesSchema = z2.record(z2.string(), z2.record(z2.string(), ProviderScalarS
   }
 });
 var StaticSchema = z2.record(z2.string(), ProviderScalarSchema);
+var ArraysSchema = z2.record(z2.string().min(1), z2.array(z2.string()).min(1)).superRefine((arrays, ctx) => {
+  for (const [name, keys] of Object.entries(arrays)) {
+    for (const key of keys) {
+      if (!ARRAY_KEY_SET.has(key)) {
+        ctx.addIssue({
+          code: "custom",
+          path: [name],
+          message: `inputSpec.arrays "${name}" has unknown image key "${key}"`
+        });
+      }
+    }
+  }
+});
 var InputSpecSchema = z2.strictObject({
   fields: FieldsSchema,
   values: ValuesSchema.optional(),
   static: StaticSchema.optional(),
+  arrays: ArraysSchema.optional(),
   editEndpoint: z2.string().min(1).optional()
 }).superRefine((spec, ctx) => {
-  const mappedNames = new Set(Object.values(spec.fields));
+  const mappedNames = /* @__PURE__ */ new Set([
+    ...Object.values(spec.fields),
+    ...Object.keys(spec.arrays ?? {})
+  ]);
   for (const key of Object.keys(spec.static ?? {})) {
     if (mappedNames.has(key)) {
       ctx.addIssue({
@@ -2311,6 +2359,9 @@ var InputSpecSchema = z2.strictObject({
 function isInputSpecKey(key) {
   return ALL_KEY_SET.has(key);
 }
+function isArrayInputSpecKey(key) {
+  return ARRAY_KEY_SET.has(key);
+}
 function validateInputSpec(raw, row) {
   const parsed = InputSpecSchema.safeParse(raw);
   if (!parsed.success) {
@@ -2389,14 +2440,54 @@ function validateInputSpec(raw, row) {
       });
     }
   }
+  const arrayOwner = /* @__PURE__ */ new Map();
+  for (const [name, keys] of Object.entries(spec.arrays ?? {})) {
+    const fieldKey = seenNames.get(name);
+    if (fieldKey !== void 0) {
+      issues.push({
+        path: ["arrays", name],
+        message: `inputSpec.arrays "${name}" is also the provider field of inputSpec.fields "${fieldKey}"`
+      });
+    }
+    for (const key of keys) {
+      if (allowed !== void 0 && !allowed.has(key)) {
+        issues.push({
+          path: ["arrays", name],
+          message: `inputSpec.arrays "${key}" is not supported for ${row.type} rows`
+        });
+      }
+      if (key in spec.fields) {
+        issues.push({
+          path: ["arrays", name],
+          message: `inputSpec.arrays "${key}" is already mapped in inputSpec.fields`
+        });
+      }
+      const owner = arrayOwner.get(key);
+      if (owner === void 0) arrayOwner.set(key, name);
+      else {
+        issues.push({
+          path: ["arrays", name],
+          message: owner === name ? `inputSpec.arrays "${name}" lists "${key}" twice` : `inputSpec.arrays "${owner}" and "${name}" both take "${key}"`
+        });
+      }
+    }
+  }
   const supports = row.constraints.supports ?? [];
   for (const key of INPUT_SPEC_SUPPORTS_KEYS) {
-    if (key in spec.fields && !supports.includes(key)) {
+    if (supports.includes(key)) continue;
+    if (key in spec.fields) {
       issues.push({
         path: ["fields", key],
         message: `inputSpec.fields "${key}" needs "${key}" in constraints.supports`
       });
     }
+    const arrayName = arrayOwner.get(key);
+    if (arrayName !== void 0) {
+      issues.push({
+        path: ["arrays", arrayName],
+        message: `inputSpec.arrays "${key}" needs "${key}" in constraints.supports`
+      });
+    }
   }
   if (issues.length > 0) return { ok: false, issues };
   return { ok: true, spec };
@@ -2445,6 +2536,10 @@ var AspectRatioSchema = z3.enum([
   "21:9",
   "9:21",
   "97:30",
+  "1:4",
+  "4:1",
+  "1:8",
+  "8:1",
   "adaptive"
 ]);
 var ImageSizeSchema = z3.enum(["512", "1K", "2K", "4K"]);
@@ -2614,6 +2709,7 @@ var CatalogCapabilitiesSchema = z3.looseObject({
   acceptsSourceVideo: z3.boolean().optional(),
   videoEdit: z3.boolean().optional(),
   framesExcludeReferences: z3.boolean().optional(),
+  sourceVideoExcludesFrames: z3.boolean().optional(),
   lastFrameRequiresFirstFrame: z3.boolean().optional(),
   reasoning: z3.boolean().optional(),
   vision: z3.boolean().optional(),
@@ -4141,6 +4237,17 @@ function applyInputSpec(spec, req, refToUrl2, opts) {
     const mapped = spec.values?.[canonical]?.[String(value)];
     out[providerField] = mapped ?? serialised;
   }
+  for (const [providerField, keys] of Object.entries(spec.arrays ?? {})) {
+    const urls = [];
+    for (const key of keys) {
+      if (!isArrayInputSpecKey(key)) continue;
+      const value = canonicalValue(req, key);
+      if (value === void 0) continue;
+      if (Array.isArray(value)) urls.push(...value.map(refToUrl2));
+      else if (isMediaRef(value)) urls.push(refToUrl2(value));
+    }
+    if (urls.length > 0) out[providerField] = urls;
+  }
   for (const [key, value] of Object.entries(spec.static ?? {})) {
     out[key] = value;
   }
@@ -8262,6 +8369,26 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
     });
     delete merged.sourceVideo;
   }
+  const imageInputs = ["firstFrame", "lastFrame", "referenceImages"].filter(
+    (field) => field === "referenceImages" ? compact(merged.referenceImages).length > 0 : merged[field] !== void 0
+  );
+  if (merged.sourceVideo !== void 0 && imageInputs.length > 0 && model.capabilities?.sourceVideoExcludesFrames) {
+    if (onUnsupported === "error") {
+      throw new ModelKitError({
+        code: "unsupported_option",
+        message: `${model.id} takes a source video or images, not both`,
+        model: model.id
+      });
+    }
+    for (const field of imageInputs) {
+      warnings.push({
+        code: "option_ignored",
+        field,
+        message: `${model.id} takes a source video or images, not both; ${field} dropped`
+      });
+      delete merged[field];
+    }
+  }
   if (merged.lastFrame !== void 0 && merged.firstFrame === void 0 && model.capabilities?.lastFrameRequiresFirstFrame) {
     if (onUnsupported === "error") {
       throw new ModelKitError({
@@ -9378,6 +9505,7 @@ var describe = (modelId) => createDefaultModelKit().describe(modelId);
 export {
   ADAPTER_IDS,
   ADAPTER_KEYS,
+  ARRAY_INPUT_SPEC_KEYS,
   AUDIO_INPUT_SPEC_KEYS,
   Catalog,
   CatalogModelSchema,
@@ -9420,6 +9548,7 @@ export {
   googleImageAdapter,
   hasAwsCredentials,
   higgsfieldAdapter,
+  isArrayInputSpecKey,
   isInputSpecKey,
   isListed,
   isListedOn,