@shop-os/modelkit 0.7.1 → 0.7.3

package/dist/index.d.cts

@@ -1769,6 +1769,15 @@ declare function resolveImageOptions(model: CatalogModel, request: ImageRequest,
 declare function resolveVideoOptions(model: CatalogModel, request: VideoRequest, onUnsupported?: OnUnsupported): ResolveResult<VideoRequest>;
 declare function resolveAudioOptions(model: CatalogModel, request: AudioRequest, onUnsupported?: OnUnsupported): ResolveResult<AudioRequest>;
 
+/**
+ * A row whose aspect ratio is the frame it renders its source into (Ray 3.2 Reframe): it
+ * requires a source video and lists ratios without "adaptive", so the ratio is a target. Every
+ * other edit row keeps the source's framing.
+ */
+declare function takesTargetAspectRatio(model: Pick<CatalogModel, "constraints">): boolean;
+/** Whether a Luma reframe asks for a vertical target at 1080p, which Luma does not render. */
+declare function lumaReframeTooTall(req: Pick<VideoRequest, "aspectRatio" | "resolution">): boolean;
+
 /** `generateSpeech` on a `speech` row. */
 declare function resolveSpeechOptions(model: CatalogModel, request: SpeechRequest, onUnsupported?: OnUnsupported): ResolveResult<SpeechRequest>;
 /** `changeVoice` on a `voice-change` row. */
@@ -2142,4 +2151,4 @@ 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 VoiceChangeRequest, type VoiceGender, type VoiceOption, 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, lumaAdapter, 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, 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 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 VoiceChangeRequest, type VoiceGender, type VoiceOption, 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, 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, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.d.ts

@@ -1769,6 +1769,15 @@ declare function resolveImageOptions(model: CatalogModel, request: ImageRequest,
 declare function resolveVideoOptions(model: CatalogModel, request: VideoRequest, onUnsupported?: OnUnsupported): ResolveResult<VideoRequest>;
 declare function resolveAudioOptions(model: CatalogModel, request: AudioRequest, onUnsupported?: OnUnsupported): ResolveResult<AudioRequest>;
 
+/**
+ * A row whose aspect ratio is the frame it renders its source into (Ray 3.2 Reframe): it
+ * requires a source video and lists ratios without "adaptive", so the ratio is a target. Every
+ * other edit row keeps the source's framing.
+ */
+declare function takesTargetAspectRatio(model: Pick<CatalogModel, "constraints">): boolean;
+/** Whether a Luma reframe asks for a vertical target at 1080p, which Luma does not render. */
+declare function lumaReframeTooTall(req: Pick<VideoRequest, "aspectRatio" | "resolution">): boolean;
+
 /** `generateSpeech` on a `speech` row. */
 declare function resolveSpeechOptions(model: CatalogModel, request: SpeechRequest, onUnsupported?: OnUnsupported): ResolveResult<SpeechRequest>;
 /** `changeVoice` on a `voice-change` row. */
@@ -2142,4 +2151,4 @@ 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 VoiceChangeRequest, type VoiceGender, type VoiceOption, 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, lumaAdapter, 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, 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 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 VoiceChangeRequest, type VoiceGender, type VoiceOption, 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, 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, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.js

@@ -6350,6 +6350,57 @@ function applyXaiVideoRules(model, req, onUnsupported, warnings) {
   });
   req.resolution = XAI_REFERENCE_MAX_RESOLUTION;
 }
+function takesTargetAspectRatio(model) {
+  const ratios = model.constraints.aspectRatios ?? [];
+  return (model.constraints.requires ?? []).includes("sourceVideo") && ratios.length > 0 && !ratios.includes("adaptive");
+}
+var LUMA_REFRAME_VERTICAL_MAX_RESOLUTION = "720p";
+function lumaReframeTooTall(req) {
+  const vertical = req.aspectRatio === "9:16" || req.aspectRatio === "3:4";
+  return vertical && req.resolution === "1080p";
+}
+function applyLumaVideoRules(model, req, onUnsupported, warnings) {
+  if (model.adapter !== "luma" || !takesTargetAspectRatio(model)) return;
+  if (req.aspectRatio === void 0 || req.aspectRatio === "adaptive") {
+    const listed2 = model.constraints.aspectRatios ?? [];
+    const fallback = model.defaults.aspectRatio !== void 0 && model.defaults.aspectRatio !== "adaptive" ? model.defaults.aspectRatio : listed2[0];
+    const message2 = `${model.id} needs a target aspect ratio: ${listed2.join(", ")}`;
+    if (onUnsupported === "error" || fallback === void 0) {
+      throw new ModelKitError({
+        code: "unsupported_option",
+        message: message2,
+        model: model.id,
+        retryable: false
+      });
+    }
+    warnings.push({
+      code: "ratio_coerced",
+      field: "aspectRatio",
+      from: req.aspectRatio ?? "adaptive",
+      to: fallback,
+      message: message2
+    });
+    req.aspectRatio = fallback;
+  }
+  if (!lumaReframeTooTall(req)) return;
+  const message = `${model.id} renders 9:16 and 3:4 reframes at up to ${LUMA_REFRAME_VERTICAL_MAX_RESOLUTION}`;
+  if (onUnsupported === "error") {
+    throw new ModelKitError({
+      code: "unsupported_option",
+      message,
+      model: model.id,
+      retryable: false
+    });
+  }
+  warnings.push({
+    code: "resolution_coerced",
+    field: "resolution",
+    from: req.resolution,
+    to: LUMA_REFRAME_VERTICAL_MAX_RESOLUTION,
+    message
+  });
+  req.resolution = LUMA_REFRAME_VERTICAL_MAX_RESOLUTION;
+}
 var MINIMAX_H3_INSERT = "minimax/h3-max/insert-video";
 var INSERT_MIN = 1.625;
 var INSERT_MAX = 60;
@@ -13621,24 +13672,24 @@ import { z as z25 } from "zod";
 function mp4DurationSeconds(bytes) {
   const moov = findBox(bytes, 0, bytes.length, "moov");
   if (!moov) return void 0;
-  const mvhd = findBox(bytes, moov.start + 8, moov.end, "mvhd");
+  const mvhd = findBox(bytes, moov.body, moov.end, "mvhd");
   if (!mvhd) return void 0;
   const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);
-  const size = mvhd.end - mvhd.start;
-  if (size < 8 + 4) return void 0;
-  const version = view.getUint8(mvhd.start + 8);
+  const size = mvhd.end - mvhd.body;
+  if (size < 4) return void 0;
+  const version = view.getUint8(mvhd.body);
   if (version === 0) {
-    if (size < 8 + 4 + 4 + 4 + 4 + 4) return void 0;
-    const timescale = view.getUint32(mvhd.start + 8 + 4 + 4 + 4);
-    const duration = view.getUint32(mvhd.start + 8 + 4 + 4 + 4 + 4);
+    if (size < 4 + 4 + 4 + 4 + 4) return void 0;
+    const timescale = view.getUint32(mvhd.body + 4 + 4 + 4);
+    const duration = view.getUint32(mvhd.body + 4 + 4 + 4 + 4);
     if (timescale === 0) return void 0;
     return duration / timescale;
   }
   if (version === 1) {
-    if (size < 8 + 4 + 8 + 8 + 4 + 8) return void 0;
-    const timescale = view.getUint32(mvhd.start + 8 + 4 + 8 + 8);
-    const high = view.getUint32(mvhd.start + 8 + 4 + 8 + 8 + 4);
-    const low = view.getUint32(mvhd.start + 8 + 4 + 8 + 8 + 4 + 4);
+    if (size < 4 + 8 + 8 + 4 + 8) return void 0;
+    const timescale = view.getUint32(mvhd.body + 4 + 8 + 8);
+    const high = view.getUint32(mvhd.body + 4 + 8 + 8 + 4);
+    const low = view.getUint32(mvhd.body + 4 + 8 + 8 + 4 + 4);
     if (timescale === 0) return void 0;
     return (high * 2 ** 32 + low) / timescale;
   }
@@ -13648,15 +13699,23 @@ function findBox(bytes, from, to, type) {
   const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);
   let offset = from;
   while (offset + 8 <= to) {
-    const size = view.getUint32(offset);
+    let size = view.getUint32(offset);
+    let header = 8;
     const name = String.fromCharCode(
       view.getUint8(offset + 4),
       view.getUint8(offset + 5),
       view.getUint8(offset + 6),
       view.getUint8(offset + 7)
     );
-    if (size < 8 || offset + size > to) return void 0;
-    if (name === type) return { start: offset, end: offset + size };
+    if (size === 1) {
+      if (offset + 16 > to) return void 0;
+      size = view.getUint32(offset + 8) * 2 ** 32 + view.getUint32(offset + 12);
+      header = 16;
+    } else if (size === 0) {
+      size = to - offset;
+    }
+    if (size < header || offset + size > to) return void 0;
+    if (name === type) return { body: offset + header, end: offset + size };
     offset += size;
   }
   return void 0;
@@ -13669,6 +13728,7 @@ var IMAGE_TIMEOUT_MS = 5 * 6e4;
 var VIDEO_POLL_INTERVAL_MS = 5e3;
 var VIDEO_TIMEOUT_MS = 20 * 6e4;
 var PROMPT_MAX_CHARS = 6e3;
+var OUTPUT_URL_TTL_MS = 55 * 6e4;
 var LUMA_IMAGE_MODEL_IDS = /* @__PURE__ */ new Set(["uni-1", "uni-1-max"]);
 var LUMA_VIDEO_MODEL_ID = "ray-3.2";
 var LUMA_VIDEO_MODES = /* @__PURE__ */ new Map([
@@ -13687,6 +13747,9 @@ function lumaRowRefusal(row2) {
   }
   return `luma has no ${row2.type} rows`;
 }
+function lumaModel(ctx) {
+  return ctx.model.providerModelId.toLowerCase();
+}
 function lumaBase(ctx) {
   return ctx.endpoints?.luma ?? DEFAULT_BASE_URL;
 }
@@ -13729,8 +13792,23 @@ function imageRef(ref) {
   if ("url" in ref) return { url: ref.url };
   return { data: mediaRefToBase64(ref.bytes), media_type: ref.mime };
 }
+var VIDEO_TYPES_BY_EXTENSION = /* @__PURE__ */ new Map([
+  ["mp4", "video/mp4"],
+  ["m4v", "video/mp4"],
+  ["mov", "video/quicktime"],
+  ["webm", "video/webm"]
+]);
+function videoUrlType(url) {
+  let path = url;
+  try {
+    path = new URL(url).pathname;
+  } catch {
+  }
+  const extension = path.slice(path.lastIndexOf(".") + 1).toLowerCase();
+  return VIDEO_TYPES_BY_EXTENSION.get(extension) ?? "video/mp4";
+}
 function videoRef(ref) {
-  if ("url" in ref) return { url: ref.url, media_type: "video/mp4" };
+  if ("url" in ref) return { url: ref.url, media_type: videoUrlType(ref.url) };
   const mediaType = ref.mime.startsWith("video/") ? ref.mime : "video/mp4";
   return { data: mediaRefToBase64(ref.bytes), media_type: mediaType };
 }
@@ -13780,16 +13858,12 @@ function lumaVideoMode(ctx) {
   }
   return mode;
 }
-function reframeTooTall(req) {
-  const vertical = req.aspectRatio === "9:16" || req.aspectRatio === "3:4";
-  return vertical && req.resolution === "1080p";
-}
 function sentResolution(req, mode) {
-  return mode === "reframe" && reframeTooTall(req) ? "720p" : req.resolution;
+  return mode === "reframe" && lumaReframeTooTall(req) ? "720p" : req.resolution;
 }
 function lumaReframeResolution(req, ctx) {
   const requested = req.resolution;
-  if (requested === void 0 || !reframeTooTall(req)) return requested;
+  if (requested === void 0 || !lumaReframeTooTall(req)) return requested;
   const message = `${ctx.model.id} renders 9:16 and 3:4 reframes at up to 720p`;
   if ((ctx.call.onUnsupported ?? "error") === "error") {
     throw new ModelKitError({
@@ -13898,14 +13972,22 @@ async function lumaRequest(ctx, url, body) {
   }
   return res;
 }
+async function parseGeneration(res) {
+  try {
+    const parsed = lumaGenerationSchema.safeParse(await res.json());
+    return parsed.success ? parsed.data : void 0;
+  } catch {
+    return "not_json";
+  }
+}
 async function submitGeneration(ctx, body) {
   const retryOptions = mediaRetryOptions(ctx.call, ctx.signal);
   const res = await withRetry(
     () => lumaRequest(ctx, `${lumaBase(ctx)}/v1/generations`, JSON.stringify(body)),
     retryOptions
   );
-  const parsed = lumaGenerationSchema.safeParse(await res.json());
-  if (!parsed.success) {
+  const generation = await parseGeneration(res);
+  if (generation === void 0 || generation === "not_json") {
     throw new ModelKitError({
       code: "provider_rejected",
       message: "Luma submit returned no generation id",
@@ -13914,13 +13996,20 @@ async function submitGeneration(ctx, body) {
       retryable: false
     });
   }
-  return parsed.data;
+  return generation;
 }
 async function readGeneration(ctx, id) {
   const res = await lumaRequest(ctx, `${lumaBase(ctx)}/v1/generations/${id}`);
-  const payload = JsonValueSchema.parse(await res.json());
-  const parsed = lumaGenerationSchema.safeParse(payload);
-  if (!parsed.success) {
+  const generation = await parseGeneration(res);
+  if (generation === "not_json") {
+    throw new ModelKitError({
+      code: "network",
+      message: "Luma returned a poll response that is not JSON",
+      provider: "luma",
+      model: ctx.model.id
+    });
+  }
+  if (generation === void 0) {
     throw new ModelKitError({
       code: "provider_rejected",
       message: "Luma returned an unreadable generation",
@@ -13929,14 +14018,12 @@ async function readGeneration(ctx, id) {
       retryable: false
     });
   }
-  return parsed.data;
+  return generation;
 }
 function outputUrl(generation, type) {
   const outputs = generation.output ?? [];
-  const match = outputs.find((item) => item.type === type && item.url);
-  if (match?.url) return match.url;
-  const fallback = outputs.find((item) => item.url);
-  return fallback?.url;
+  const match = outputs.find((item) => item.type === type && item.url) ?? outputs.find((item) => item.type === void 0 && item.url);
+  return match?.url;
 }
 async function readGenerationOnce(ctx, id) {
   try {
@@ -13975,21 +14062,22 @@ async function checkGeneration(ctx, id, type) {
   return void 0;
 }
 async function downloadBytes(ctx, url, fallback) {
-  let bytes;
-  let header;
-  try {
-    const out = await mediaRefToBytes({ url }, ctx.fetch, {
-      ...ctx.signal !== void 0 && { signal: ctx.signal },
-      provider: "luma",
-      model: ctx.model.id
-    });
-    bytes = out.bytes;
-    header = out.mime;
-  } catch (cause) {
-    throw classifyThrown("luma", cause, ctx.model.id);
-  }
+  const { bytes, mime: header } = await withRetry(
+    async () => {
+      try {
+        return await mediaRefToBytes({ url }, ctx.fetch, {
+          ...ctx.signal !== void 0 && { signal: ctx.signal },
+          provider: "luma",
+          model: ctx.model.id
+        });
+      } catch (cause) {
+        throw classifyThrown("luma", cause, ctx.model.id);
+      }
+    },
+    mediaRetryOptions(ctx.call, ctx.signal)
+  );
   const kind = fallback.split("/")[0];
-  const usable = header !== void 0 && header.startsWith(`${kind}/`);
+  const usable = header.startsWith(`${kind}/`);
   return { bytes, mime: usable ? header : fallback };
 }
 function imageUsage2(ctx) {
@@ -14004,15 +14092,12 @@ function imageUsage2(ctx) {
 }
 var GENERATE_SECONDS = 5;
 var MAX_PLAUSIBLE_SECONDS = 60;
-function billedSeconds(req, mode, bytes) {
-  if (mode === "generate") return GENERATE_SECONDS;
+function editedSeconds(req, bytes) {
   const measured = mp4DurationSeconds(bytes);
-  const source2 = req.sourceVideo;
-  const length = measured !== void 0 && measured > 0 && measured <= MAX_PLAUSIBLE_SECONDS ? measured : source2 !== void 0 && "url" in source2 ? source2.durationSeconds : void 0;
+  const length = measured !== void 0 && measured > 0 && measured <= MAX_PLAUSIBLE_SECONDS ? measured : req.sourceVideo?.durationSeconds;
   return Math.max(length ?? 0, GENERATE_SECONDS);
 }
-function videoUsage4(ctx, req, mode, bytes) {
-  const seconds3 = billedSeconds(req, mode, bytes);
+function videoUsage4(ctx, req, mode, seconds3) {
   const event = {
     model: ctx.model.id,
     provider: "luma",
@@ -14027,7 +14112,7 @@ function videoUsage4(ctx, req, mode, bytes) {
 }
 function imageBody(req, ctx, outputFormat) {
   const prompt = checkPrompt(req, ctx);
-  const model = ctx.model.providerModelId;
+  const model = lumaModel(ctx);
   const ratio = req.aspectRatio !== "adaptive" ? req.aspectRatio : void 0;
   const [source2, ...rest] = req.references ?? [];
   if (source2 === void 0) {
@@ -14055,7 +14140,7 @@ function imageBody(req, ctx, outputFormat) {
 }
 function videoBody(req, ctx, mode) {
   const prompt = checkPrompt(req, ctx);
-  const model = ctx.model.providerModelId;
+  const model = lumaModel(ctx);
   const hasReferences = (req.referenceImages ?? []).length + (req.referenceVideos ?? []).length + (req.referenceAudios ?? []).length > 0;
   if (hasReferences) {
     refuseOrDrop(
@@ -14137,7 +14222,7 @@ function videoBody(req, ctx, mode) {
 async function generateLumaImage(req, ctx) {
   throwIfAborted(ctx.signal);
   lumaKey(ctx);
-  if (!LUMA_IMAGE_MODEL_IDS.has(ctx.model.providerModelId.toLowerCase())) {
+  if (!LUMA_IMAGE_MODEL_IDS.has(lumaModel(ctx))) {
     throw new ModelKitError({
       code: "unsupported_option",
       message: `luma has no image mapping for "${ctx.model.providerModelId}"`,
@@ -14170,16 +14255,25 @@ function lumaVideoJob(ctx, req, mode, id) {
       return { status: "pending" };
     },
     async result(opts) {
-      const signal = opts?.signal ?? ctx.signal;
-      const url = await pollUntil(checkGeneration.bind(null, ctx, id, "video"), {
+      const signal = combineSignals(ctx.signal, opts?.signal);
+      const jobCtx = { ...ctx, ...signal !== void 0 && { signal } };
+      const url = await pollUntil(checkGeneration.bind(null, jobCtx, id, "video"), {
         intervalMs: opts?.pollIntervalMs ?? VIDEO_POLL_INTERVAL_MS,
         timeoutMs: opts?.timeoutMs ?? VIDEO_TIMEOUT_MS,
         ...signal !== void 0 && { signal }
       });
-      const video = await downloadBytes(ctx, url, "video/mp4");
+      if (mode === "generate") {
+        const result2 = {
+          video: { url, expiresAt: new Date(Date.now() + OUTPUT_URL_TTL_MS) },
+          usage: videoUsage4(jobCtx, req, mode, GENERATE_SECONDS),
+          raw: { id }
+        };
+        return result2;
+      }
+      const video = await downloadBytes(jobCtx, url, "video/mp4");
       const result = {
         video,
-        usage: videoUsage4(ctx, req, mode, video.bytes),
+        usage: videoUsage4(jobCtx, req, mode, editedSeconds(req, video.bytes)),
         raw: { id }
       };
       return result;
@@ -14189,7 +14283,7 @@ function lumaVideoJob(ctx, req, mode, id) {
 async function generateLumaVideo(req, ctx) {
   throwIfAborted(ctx.signal);
   lumaKey(ctx);
-  if (ctx.model.providerModelId.toLowerCase() !== LUMA_VIDEO_MODEL_ID) {
+  if (lumaModel(ctx) !== LUMA_VIDEO_MODEL_ID) {
     throw new ModelKitError({
       code: "unsupported_option",
       message: `luma has no video mapping for "${ctx.model.providerModelId}"`,
@@ -16519,6 +16613,7 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
     delete merged.previousInteractionId;
   }
   applyXaiVideoRules(model, merged, onUnsupported, warnings);
+  applyLumaVideoRules(model, merged, onUnsupported, warnings);
   checkVideoRequired(model, merged);
   applyFalVideoRules(model, merged, request.aspectRatio, warnings);
   return { request: merged, warnings };
@@ -18123,6 +18218,7 @@ export {
   liveCatalog,
   liveState,
   lumaAdapter,
+  lumaReframeTooTall,
   mergeTextProviderOptions,
   missingTierKeys,
   model3dCall,
@@ -18160,6 +18256,7 @@ export {
   sortCatalogRows,
   streamObject,
   streamText,
+  takesTargetAspectRatio,
   textUsageEvent,
   toGeneratedImage,
   validateInputSpec,