@shop-os/modelkit 0.8.0 → 0.10.0

package/dist/index.d.cts

@@ -72,6 +72,10 @@ type ModelType = "text" | "image" | "video" | "audio"
  | "lipsync"
 /** Speech to text (`transcribe`). */
  | "transcription"
+/** A reusable voice cloned from recordings (`createVoice`); its id feeds `SpeechRequest.customVoice`. */
+ | "voice-clone"
+/** Candidate voices from a description (`designVoice`); keep one with `createVoice`. */
+ | "voice-design"
 /** A prompt and/or photos of an object in, a GLB 3D model out. */
  | "model3d";
 /**
@@ -136,7 +140,7 @@ type Lifecycle = "active" | "deprecated" | "retired";
 type Surface = "cowork" | "space_builder";
 /** @deprecated use Surface[] */
 type Exposure = Surface[];
-type AdapterId = "ai-sdk-text" | "google-image" | "openai-image" | "google-genai" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "xai-video" | "luma";
+type AdapterId = "ai-sdk-text" | "google-image" | "openai-image" | "google-genai" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "xai-video" | "luma" | "fish" | "elevenlabs";
 /** Gemini image safety threshold, per category, sent only when the caller opts in. */
 type ImageSafetySetting = {
     category: string;
@@ -430,6 +434,12 @@ interface SpeechRequest {
     text: string;
     /** One of `constraints.voices[].id`; the row default when unset. */
     voice?: string;
+    /**
+     * Rows with `capabilities.customVoices`: a voice id `createVoice` returned, used instead of
+     * `voice`. It belongs to the provider account of the key set that created it, so call with
+     * the same `tier`.
+     */
+    customVoice?: string;
     /** One of `constraints.languages[].code`; the provider detects it when unset. */
     language?: string;
     /** Rows with `supports: style`: how to speak it ("warm and slow"), kept out of the script. */
@@ -438,6 +448,65 @@ interface SpeechRequest {
     outputFormat?: string;
     abortSignal?: AbortSignal;
 }
+/** One recording a voice is cloned from. */
+interface VoiceSample {
+    /** Clean, single-speaker speech. Set `durationSeconds` so `constraints.references.audioSeconds` can be checked. */
+    audio: MediaRef;
+    /** The words spoken in `audio`; the provider transcribes the clip when unset. */
+    transcript?: string;
+    /** `VoiceCandidate.signature` when `audio` is a `designVoice` candidate. */
+    designSignature?: string;
+}
+/** `createVoice` on a `voice-clone` row: a private, reusable voice. */
+interface VoiceCloneRequest {
+    model: string;
+    /** The voice's name in the provider console. */
+    title: string;
+    samples: VoiceSample[];
+    description?: string;
+    /** Provider noise removal and level normalisation before cloning; the provider default (on) when unset. */
+    enhance?: boolean;
+    abortSignal?: AbortSignal;
+}
+type VoiceCloneState = "created" | "training" | "trained" | "failed";
+interface VoiceCloneResult {
+    /** Pass as `SpeechRequest.customVoice` on a row with `capabilities.customVoices`. */
+    voiceId: string;
+    /** `trained` means usable now; unset when the provider reports a state the kit does not know. */
+    state?: VoiceCloneState;
+    usage: UsageEvent;
+    raw?: JsonValue;
+}
+/** `designVoice` on a `voice-design` row: short candidate voices from a description. */
+interface VoiceDesignRequest {
+    model: string;
+    /** The voice: age, tone, accent, pacing, context. */
+    prompt: string;
+    /** A short line every candidate reads, so they can be compared. */
+    previewText?: string;
+    /** A language hint such as `en`. */
+    language?: string;
+    /** Candidates to return, within `constraints.voiceDesign.maxCandidates`; the row default when unset. */
+    count?: number;
+    seed?: number;
+    abortSignal?: AbortSignal;
+}
+interface VoiceCandidate {
+    audio: MediaOutput;
+    durationSeconds: number;
+    /** Send with this audio in `VoiceSample.designSignature` to keep the voice. */
+    signature: string;
+}
+interface VoiceDesignResult {
+    candidates: VoiceCandidate[];
+    usage: UsageEvent;
+}
+/** `deleteVoice` on the `voice-clone` row that created the voice, with the same `tier`. */
+interface VoiceDeleteRequest {
+    model: string;
+    voiceId: string;
+    abortSignal?: AbortSignal;
+}
 /** `changeVoice` on a `voice-change` row: the recording re-spoken in another voice. */
 interface VoiceChangeRequest {
     model: string;
@@ -673,6 +742,11 @@ interface ModelConstraints {
     ratioIgnoredWith?: ("firstFrame" | "sourceVideo")[];
     /** Speech and voice-change rows: the voices offered; `defaults.voice` is one of their ids. */
     voices?: VoiceOption[];
+    /** `voice-design` rows: how many candidates one call returns at most, and the preview line's length cap. */
+    voiceDesign?: {
+        maxCandidates: number;
+        previewMaxChars: number;
+    };
     /** Speech, transcription and dubbing (target) rows: the languages taken. */
     languages?: LanguageOption[];
     /** Audio-output rows whose provider lets the caller pick a file format; `defaults.outputFormat` is one. */
@@ -756,6 +830,8 @@ interface CatalogDisplay {
 interface CatalogCapabilities {
     /** Video rows: takes an existing video as the thing to edit or extend. */
     acceptsSourceVideo?: boolean;
+    /** Speech rows: take `SpeechRequest.customVoice`, a voice the same provider cloned. */
+    customVoices?: boolean;
     /**
      * Video rows: offered for prompt-driven edits of a source clip (`taskType: "edit"`). Absent
      * means the caller falls back to `acceptsSourceVideo`; `false` opts a row out.
@@ -904,7 +980,7 @@ interface UsageEvent {
     latencyMs?: number;
     raw?: unknown;
 }
-type ProviderKeyName = "openai" | "anthropic" | "google" | "xai" | "perplexity" | "deepseek" | "moonshot" | "minimax" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "luma" | "cloudflare";
+type ProviderKeyName = "openai" | "anthropic" | "google" | "xai" | "perplexity" | "deepseek" | "moonshot" | "minimax" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "luma" | "fish" | "elevenlabs" | "cloudflare";
 type ProviderKeys = Partial<Record<ProviderKeyName, string>> & {
     /** Higgsfield needs the pair; both switch together. */
     higgsfieldSecret?: string;
@@ -955,6 +1031,10 @@ interface AdapterEndpoints {
     ark?: string;
     /** Luma Agents API. Defaults to https://agents.lumalabs.ai. */
     luma?: string;
+    /** Fish Audio API. Defaults to https://api.fish.audio. */
+    fish?: string;
+    /** ElevenLabs API. Defaults to https://api.elevenlabs.io. */
+    elevenlabs?: string;
 }
 interface ModelKitConfig {
     keys: {
@@ -1055,7 +1135,7 @@ declare function isListed(row: CatalogModel): boolean;
  */
 declare const MODELS: CatalogModel[];
 
-declare const ADAPTER_IDS: readonly ["ai-sdk-text", "google-image", "openai-image", "google-genai", "ark", "fal", "replicate", "prodia", "higgsfield", "xai-video", "luma"];
+declare const ADAPTER_IDS: readonly ["ai-sdk-text", "google-image", "openai-image", "google-genai", "ark", "fal", "replicate", "prodia", "higgsfield", "xai-video", "luma", "fish", "elevenlabs"];
 declare const SurfaceSchema: z.ZodEnum<{
     cowork: "cowork";
     space_builder: "space_builder";
@@ -1090,6 +1170,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         dubbing: "dubbing";
         lipsync: "lipsync";
         transcription: "transcription";
+        "voice-clone": "voice-clone";
+        "voice-design": "voice-design";
         model3d: "model3d";
     }>;
     adapter: z.ZodEnum<{
@@ -1104,6 +1186,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         higgsfield: "higgsfield";
         "xai-video": "xai-video";
         luma: "luma";
+        fish: "fish";
+        elevenlabs: "elevenlabs";
     }>;
     providerModelId: z.ZodString;
     lifecycle: z.ZodEnum<{
@@ -1237,6 +1321,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             description: z.ZodOptional<z.ZodString>;
             previewUrl: z.ZodOptional<z.ZodURL>;
         }, z.core.$loose>>>;
+        voiceDesign: z.ZodOptional<z.ZodObject<{
+            maxCandidates: z.ZodNumber;
+            previewMaxChars: z.ZodNumber;
+        }, z.core.$strip>>;
         languages: z.ZodOptional<z.ZodArray<z.ZodObject<{
             code: z.ZodString;
             name: z.ZodString;
@@ -1541,6 +1629,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
     }, z.core.$loose>>;
     capabilities: z.ZodOptional<z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
+        customVoices: z.ZodOptional<z.ZodBoolean>;
         videoEdit: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
         sourceVideoExcludesFrames: z.ZodOptional<z.ZodBoolean>;
@@ -1664,6 +1753,9 @@ interface Adapter {
     lipsync?(req: LipsyncRequest, ctx: AdapterContext): Promise<VideoJob>;
     dub?(req: DubbingRequest, ctx: AdapterContext): Promise<DubbingOutcome>;
     transcribe?(req: TranscriptionRequest, ctx: AdapterContext): Promise<TranscriptionResult>;
+    createVoice?(req: VoiceCloneRequest, ctx: AdapterContext): Promise<VoiceCloneResult>;
+    designVoice?(req: VoiceDesignRequest, ctx: AdapterContext): Promise<VoiceDesignResult>;
+    deleteVoice?(req: VoiceDeleteRequest, ctx: AdapterContext): Promise<void>;
     /** Video tools run as provider jobs the caller tracks; see ModelKit.submitVideoTool. */
     submitVideoTool?(req: VideoToolRequest, ctx: AdapterContext, opts: {
         webhookUrl?: string;
@@ -1789,6 +1881,11 @@ declare function resolveDubbingOptions(model: CatalogModel, request: DubbingRequ
 /** `transcribe` on a `transcription` row. */
 declare function resolveTranscriptionOptions(model: CatalogModel, request: TranscriptionRequest, onUnsupported?: OnUnsupported): ResolveResult<TranscriptionRequest>;
 
+/** `createVoice` on a `voice-clone` row. */
+declare function resolveVoiceCloneOptions(model: CatalogModel, request: VoiceCloneRequest): ResolveResult<VoiceCloneRequest>;
+/** `designVoice` on a `voice-design` row; a count over the row cap is clamped when coercing. */
+declare function resolveVoiceDesignOptions(model: CatalogModel, request: VoiceDesignRequest, onUnsupported?: OnUnsupported): ResolveResult<VoiceDesignRequest>;
+
 /** Which key names an adapter needs. All of them must be set on a tier for that tier to win. */
 declare const ADAPTER_KEYS: Record<AdapterId, ProviderKeyName[]>;
 /** The required names the given non-pro tier is missing. */
@@ -1847,6 +1944,16 @@ interface AudioNodeMethods {
     transcribe(req: TranscriptionRequest, call?: CallContext): Promise<TranscriptionResult>;
     /** Dubs a video or an audio file into `targetLanguage`; the result is the same kind. */
     dub(req: DubbingRequest, call?: MediaJobCallContext): Promise<DubbingResult>;
+    /**
+     * Clones a private voice from recordings on a `voice-clone` row. The voice lives in the
+     * provider account of the call's key set (`tier`); speak with it through
+     * `generateSpeech({ customVoice })` on the same tier.
+     */
+    createVoice(req: VoiceCloneRequest, call?: CallContext): Promise<VoiceCloneResult>;
+    /** Candidate voices from a description on a `voice-design` row; keep one with `createVoice`. */
+    designVoice(req: VoiceDesignRequest, call?: CallContext): Promise<VoiceDesignResult>;
+    /** Deletes a voice `createVoice` made, on the same row and tier. A voice this key set does not have is `provider_rejected` (404). */
+    deleteVoice(req: VoiceDeleteRequest, call?: CallContext): Promise<void>;
 }
 
 /**
@@ -1992,6 +2099,10 @@ declare const arkAdapter: Adapter;
 
 declare const falAdapter: Adapter;
 
+declare const elevenLabsAdapter: Adapter;
+
+declare const fishAdapter: Adapter;
+
 declare const googleGenaiAdapter: Adapter;
 
 declare const googleImageAdapter: Adapter;
@@ -2151,4 +2262,4 @@ 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, 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 };
+export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };

package/dist/index.d.ts

@@ -72,6 +72,10 @@ type ModelType = "text" | "image" | "video" | "audio"
  | "lipsync"
 /** Speech to text (`transcribe`). */
  | "transcription"
+/** A reusable voice cloned from recordings (`createVoice`); its id feeds `SpeechRequest.customVoice`. */
+ | "voice-clone"
+/** Candidate voices from a description (`designVoice`); keep one with `createVoice`. */
+ | "voice-design"
 /** A prompt and/or photos of an object in, a GLB 3D model out. */
  | "model3d";
 /**
@@ -136,7 +140,7 @@ type Lifecycle = "active" | "deprecated" | "retired";
 type Surface = "cowork" | "space_builder";
 /** @deprecated use Surface[] */
 type Exposure = Surface[];
-type AdapterId = "ai-sdk-text" | "google-image" | "openai-image" | "google-genai" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "xai-video" | "luma";
+type AdapterId = "ai-sdk-text" | "google-image" | "openai-image" | "google-genai" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "xai-video" | "luma" | "fish" | "elevenlabs";
 /** Gemini image safety threshold, per category, sent only when the caller opts in. */
 type ImageSafetySetting = {
     category: string;
@@ -430,6 +434,12 @@ interface SpeechRequest {
     text: string;
     /** One of `constraints.voices[].id`; the row default when unset. */
     voice?: string;
+    /**
+     * Rows with `capabilities.customVoices`: a voice id `createVoice` returned, used instead of
+     * `voice`. It belongs to the provider account of the key set that created it, so call with
+     * the same `tier`.
+     */
+    customVoice?: string;
     /** One of `constraints.languages[].code`; the provider detects it when unset. */
     language?: string;
     /** Rows with `supports: style`: how to speak it ("warm and slow"), kept out of the script. */
@@ -438,6 +448,65 @@ interface SpeechRequest {
     outputFormat?: string;
     abortSignal?: AbortSignal;
 }
+/** One recording a voice is cloned from. */
+interface VoiceSample {
+    /** Clean, single-speaker speech. Set `durationSeconds` so `constraints.references.audioSeconds` can be checked. */
+    audio: MediaRef;
+    /** The words spoken in `audio`; the provider transcribes the clip when unset. */
+    transcript?: string;
+    /** `VoiceCandidate.signature` when `audio` is a `designVoice` candidate. */
+    designSignature?: string;
+}
+/** `createVoice` on a `voice-clone` row: a private, reusable voice. */
+interface VoiceCloneRequest {
+    model: string;
+    /** The voice's name in the provider console. */
+    title: string;
+    samples: VoiceSample[];
+    description?: string;
+    /** Provider noise removal and level normalisation before cloning; the provider default (on) when unset. */
+    enhance?: boolean;
+    abortSignal?: AbortSignal;
+}
+type VoiceCloneState = "created" | "training" | "trained" | "failed";
+interface VoiceCloneResult {
+    /** Pass as `SpeechRequest.customVoice` on a row with `capabilities.customVoices`. */
+    voiceId: string;
+    /** `trained` means usable now; unset when the provider reports a state the kit does not know. */
+    state?: VoiceCloneState;
+    usage: UsageEvent;
+    raw?: JsonValue;
+}
+/** `designVoice` on a `voice-design` row: short candidate voices from a description. */
+interface VoiceDesignRequest {
+    model: string;
+    /** The voice: age, tone, accent, pacing, context. */
+    prompt: string;
+    /** A short line every candidate reads, so they can be compared. */
+    previewText?: string;
+    /** A language hint such as `en`. */
+    language?: string;
+    /** Candidates to return, within `constraints.voiceDesign.maxCandidates`; the row default when unset. */
+    count?: number;
+    seed?: number;
+    abortSignal?: AbortSignal;
+}
+interface VoiceCandidate {
+    audio: MediaOutput;
+    durationSeconds: number;
+    /** Send with this audio in `VoiceSample.designSignature` to keep the voice. */
+    signature: string;
+}
+interface VoiceDesignResult {
+    candidates: VoiceCandidate[];
+    usage: UsageEvent;
+}
+/** `deleteVoice` on the `voice-clone` row that created the voice, with the same `tier`. */
+interface VoiceDeleteRequest {
+    model: string;
+    voiceId: string;
+    abortSignal?: AbortSignal;
+}
 /** `changeVoice` on a `voice-change` row: the recording re-spoken in another voice. */
 interface VoiceChangeRequest {
     model: string;
@@ -673,6 +742,11 @@ interface ModelConstraints {
     ratioIgnoredWith?: ("firstFrame" | "sourceVideo")[];
     /** Speech and voice-change rows: the voices offered; `defaults.voice` is one of their ids. */
     voices?: VoiceOption[];
+    /** `voice-design` rows: how many candidates one call returns at most, and the preview line's length cap. */
+    voiceDesign?: {
+        maxCandidates: number;
+        previewMaxChars: number;
+    };
     /** Speech, transcription and dubbing (target) rows: the languages taken. */
     languages?: LanguageOption[];
     /** Audio-output rows whose provider lets the caller pick a file format; `defaults.outputFormat` is one. */
@@ -756,6 +830,8 @@ interface CatalogDisplay {
 interface CatalogCapabilities {
     /** Video rows: takes an existing video as the thing to edit or extend. */
     acceptsSourceVideo?: boolean;
+    /** Speech rows: take `SpeechRequest.customVoice`, a voice the same provider cloned. */
+    customVoices?: boolean;
     /**
      * Video rows: offered for prompt-driven edits of a source clip (`taskType: "edit"`). Absent
      * means the caller falls back to `acceptsSourceVideo`; `false` opts a row out.
@@ -904,7 +980,7 @@ interface UsageEvent {
     latencyMs?: number;
     raw?: unknown;
 }
-type ProviderKeyName = "openai" | "anthropic" | "google" | "xai" | "perplexity" | "deepseek" | "moonshot" | "minimax" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "luma" | "cloudflare";
+type ProviderKeyName = "openai" | "anthropic" | "google" | "xai" | "perplexity" | "deepseek" | "moonshot" | "minimax" | "ark" | "fal" | "replicate" | "prodia" | "higgsfield" | "luma" | "fish" | "elevenlabs" | "cloudflare";
 type ProviderKeys = Partial<Record<ProviderKeyName, string>> & {
     /** Higgsfield needs the pair; both switch together. */
     higgsfieldSecret?: string;
@@ -955,6 +1031,10 @@ interface AdapterEndpoints {
     ark?: string;
     /** Luma Agents API. Defaults to https://agents.lumalabs.ai. */
     luma?: string;
+    /** Fish Audio API. Defaults to https://api.fish.audio. */
+    fish?: string;
+    /** ElevenLabs API. Defaults to https://api.elevenlabs.io. */
+    elevenlabs?: string;
 }
 interface ModelKitConfig {
     keys: {
@@ -1055,7 +1135,7 @@ declare function isListed(row: CatalogModel): boolean;
  */
 declare const MODELS: CatalogModel[];
 
-declare const ADAPTER_IDS: readonly ["ai-sdk-text", "google-image", "openai-image", "google-genai", "ark", "fal", "replicate", "prodia", "higgsfield", "xai-video", "luma"];
+declare const ADAPTER_IDS: readonly ["ai-sdk-text", "google-image", "openai-image", "google-genai", "ark", "fal", "replicate", "prodia", "higgsfield", "xai-video", "luma", "fish", "elevenlabs"];
 declare const SurfaceSchema: z.ZodEnum<{
     cowork: "cowork";
     space_builder: "space_builder";
@@ -1090,6 +1170,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         dubbing: "dubbing";
         lipsync: "lipsync";
         transcription: "transcription";
+        "voice-clone": "voice-clone";
+        "voice-design": "voice-design";
         model3d: "model3d";
     }>;
     adapter: z.ZodEnum<{
@@ -1104,6 +1186,8 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
         higgsfield: "higgsfield";
         "xai-video": "xai-video";
         luma: "luma";
+        fish: "fish";
+        elevenlabs: "elevenlabs";
     }>;
     providerModelId: z.ZodString;
     lifecycle: z.ZodEnum<{
@@ -1237,6 +1321,10 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
             description: z.ZodOptional<z.ZodString>;
             previewUrl: z.ZodOptional<z.ZodURL>;
         }, z.core.$loose>>>;
+        voiceDesign: z.ZodOptional<z.ZodObject<{
+            maxCandidates: z.ZodNumber;
+            previewMaxChars: z.ZodNumber;
+        }, z.core.$strip>>;
         languages: z.ZodOptional<z.ZodArray<z.ZodObject<{
             code: z.ZodString;
             name: z.ZodString;
@@ -1541,6 +1629,7 @@ declare const CatalogModelSchemaBase: z.ZodObject<{
     }, z.core.$loose>>;
     capabilities: z.ZodOptional<z.ZodObject<{
         acceptsSourceVideo: z.ZodOptional<z.ZodBoolean>;
+        customVoices: z.ZodOptional<z.ZodBoolean>;
         videoEdit: z.ZodOptional<z.ZodBoolean>;
         framesExcludeReferences: z.ZodOptional<z.ZodBoolean>;
         sourceVideoExcludesFrames: z.ZodOptional<z.ZodBoolean>;
@@ -1664,6 +1753,9 @@ interface Adapter {
     lipsync?(req: LipsyncRequest, ctx: AdapterContext): Promise<VideoJob>;
     dub?(req: DubbingRequest, ctx: AdapterContext): Promise<DubbingOutcome>;
     transcribe?(req: TranscriptionRequest, ctx: AdapterContext): Promise<TranscriptionResult>;
+    createVoice?(req: VoiceCloneRequest, ctx: AdapterContext): Promise<VoiceCloneResult>;
+    designVoice?(req: VoiceDesignRequest, ctx: AdapterContext): Promise<VoiceDesignResult>;
+    deleteVoice?(req: VoiceDeleteRequest, ctx: AdapterContext): Promise<void>;
     /** Video tools run as provider jobs the caller tracks; see ModelKit.submitVideoTool. */
     submitVideoTool?(req: VideoToolRequest, ctx: AdapterContext, opts: {
         webhookUrl?: string;
@@ -1789,6 +1881,11 @@ declare function resolveDubbingOptions(model: CatalogModel, request: DubbingRequ
 /** `transcribe` on a `transcription` row. */
 declare function resolveTranscriptionOptions(model: CatalogModel, request: TranscriptionRequest, onUnsupported?: OnUnsupported): ResolveResult<TranscriptionRequest>;
 
+/** `createVoice` on a `voice-clone` row. */
+declare function resolveVoiceCloneOptions(model: CatalogModel, request: VoiceCloneRequest): ResolveResult<VoiceCloneRequest>;
+/** `designVoice` on a `voice-design` row; a count over the row cap is clamped when coercing. */
+declare function resolveVoiceDesignOptions(model: CatalogModel, request: VoiceDesignRequest, onUnsupported?: OnUnsupported): ResolveResult<VoiceDesignRequest>;
+
 /** Which key names an adapter needs. All of them must be set on a tier for that tier to win. */
 declare const ADAPTER_KEYS: Record<AdapterId, ProviderKeyName[]>;
 /** The required names the given non-pro tier is missing. */
@@ -1847,6 +1944,16 @@ interface AudioNodeMethods {
     transcribe(req: TranscriptionRequest, call?: CallContext): Promise<TranscriptionResult>;
     /** Dubs a video or an audio file into `targetLanguage`; the result is the same kind. */
     dub(req: DubbingRequest, call?: MediaJobCallContext): Promise<DubbingResult>;
+    /**
+     * Clones a private voice from recordings on a `voice-clone` row. The voice lives in the
+     * provider account of the call's key set (`tier`); speak with it through
+     * `generateSpeech({ customVoice })` on the same tier.
+     */
+    createVoice(req: VoiceCloneRequest, call?: CallContext): Promise<VoiceCloneResult>;
+    /** Candidate voices from a description on a `voice-design` row; keep one with `createVoice`. */
+    designVoice(req: VoiceDesignRequest, call?: CallContext): Promise<VoiceDesignResult>;
+    /** Deletes a voice `createVoice` made, on the same row and tier. A voice this key set does not have is `provider_rejected` (404). */
+    deleteVoice(req: VoiceDeleteRequest, call?: CallContext): Promise<void>;
 }
 
 /**
@@ -1992,6 +2099,10 @@ declare const arkAdapter: Adapter;
 
 declare const falAdapter: Adapter;
 
+declare const elevenLabsAdapter: Adapter;
+
+declare const fishAdapter: Adapter;
+
 declare const googleGenaiAdapter: Adapter;
 
 declare const googleImageAdapter: Adapter;
@@ -2151,4 +2262,4 @@ 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, 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 };
+export { ADAPTER_IDS, ADAPTER_KEYS, ARRAY_INPUT_SPEC_KEYS, AUDIO_INPUT_SPEC_KEYS, type Adapter, type AdapterContext, type AdapterEndpoints, type AdapterId, type AdapterRegistry, type AiTextProviderOptions, type ArrayInputSpecKey, type AspectRatio, type AudioNodeMethods, type AudioOutput, type AudioRequest, type AudioResult, type CallContext, Catalog, type CatalogAuth, type CatalogCache, type CatalogCapabilities, type CatalogDisplay, type CatalogEnvelope, type CatalogModel, CatalogModelSchema, CatalogModelSchemaBase, type CatalogOrigin, CatalogSchema, type CatalogSource, type CatalogTextModel, DEFAULT_ADAPTERS, type DataOnlyRow, type DataOnlyVerdict, type DefaultObjectOutput, type DubbingOutcome, type DubbingRequest, type DubbingResult, type Exposure, ExposureSchema, type FalWebhookBody, type GeneratedImage, IMAGE_INPUT_SPEC_KEYS, INPUT_SPEC_SUPPORTS_KEYS, type ImageBytes, type ImageCallOptions, type ImageCallResult, type ImageQuality, type ImageRequest, type ImageResult, type ImageSafetySetting, type ImageSize, type InputSpec, type InputSpecDimensions, type InputSpecIssue, type InputSpecKey, InputSpecSchema, type JsonObject, JsonObjectSchema, type JsonPrimitive, type JsonValue, JsonValueSchema, type KeyTier, type LanguageModelHandle, type LanguageOption, type Lifecycle, type LipsyncRequest, type LipsyncResult, type LiveCatalogState, MODEL3D_IMAGE_SLOTS, MODEL3D_MODELS, MODELS, type MaskBox, type MaskPoint, type MediaJobCallContext, type MediaNodeDefaults, type MediaOutput, type MediaPart, type MediaRef, type Model3dCall, type Model3dConstraints, type Model3dEndpoints, type Model3dFamily, type Model3dInputRefusal, type Model3dPricedInputs, type Model3dPricedOptions, type Model3dRequest, type Model3dResult, type Model3dStatus, type Model3dSubmission, type Model3dTexture, type Model3dViews, type Model3dWired, type ModelConstraints, type ModelKit, type ModelKitConfig, ModelKitError, type ModelKitErrorCode, type ModelKitErrorInit, type ModelType, type ModelkitGenerateObjectOptions, type ModelkitGenerateTextOptions, type ModelkitStreamObjectOptions, type ModelkitStreamTextOptions, type ObjectOutput, type ObjectResult, type OnUnsupported, type OutputFormat, type PricingKind, type PrimeCatalogResult, type PromptConventions, type ProviderErrorBody, type ProviderKeyName, type ProviderKeys, type ProviderOptionValue, type ProviderOptions, type QuarantineResult, type ReferenceLimits, type ReferenceSyntax, type RequiredInput, type RequiredKeyName, type ResolveResult, type Routing, SURFACES, type SpeechRequest, type SubtitleOptions, type SubtitleTextStyle, type SupportedOption, type Surface, SurfaceSchema, type TextMethods, type TextModelInstance, type TranscriptWord, type TranscriptionRequest, type TranscriptionResult, type UpscaleOptions, type UsageEvent, type UsageModality, type UsageUnit, VIDEO_INPUT_SPEC_KEYS, VIDEO_TOOL_MODELS, type VideoCallJob, type VideoCallOptions, type VideoJob, type VideoModel, type VideoRequest, type VideoResolution, VideoResolutionSchema, type VideoResult, type VideoSecondsLimits, type VideoToolCallContext, type VideoToolConstraints, type VideoToolKind, type VideoToolOutputFormat, type VideoToolOutputSpec, type VideoToolPlan, type VideoToolRequest, type VideoToolResult, type VideoToolSource, type VideoToolStatus, type VideoToolSubmission, type VoiceCandidate, type VoiceChangeRequest, type VoiceCloneRequest, type VoiceCloneResult, type VoiceCloneState, type VoiceDeleteRequest, type VoiceDesignRequest, type VoiceDesignResult, type VoiceGender, type VoiceOption, type VoiceSample, type Warning, adapterServesType, aiSdkTextAdapter, arkAdapter, aspectRatioToSize, classifyHttpError, compareCatalogOrder, createDefaultModelKit, createModelKit, dataOnlyCapable, describe, describeThrown, elevenLabsAdapter, generateImage as experimental_generateImage, falAdapter, falWebhookBodySchema, fileCatalogCache, fishAdapter, generateAudio, generateImage, generateObject, generateText, generateVideo, googleGenaiAdapter, googleImageAdapter, hasAwsCredentials, higgsfieldAdapter, isArrayInputSpecKey, isInputSpecKey, isListed, isListedOn, isModelKitError, keysFromEnv, languageModel, liveCatalog, liveState, lumaAdapter, lumaReframeTooTall, mergeTextProviderOptions, missingTierKeys, model3dCall, model3dConstraints, model3dInputRefusal, model3dPricingSku, model3dResolution, model3dSeedFits, model3dSkus, model3dTexture, nearestAspectRatio, normalizeExposure, openaiImageAdapter, passthroughTextModel, planVideoTool, pollUntil, prepareMediaPart, primeCatalog, prodiaAdapter, readErrorBody, replicateAdapter, resetLiveCatalog, resolutionTier, resolveAudioOptions, resolveDubbingOptions, resolveImageOptions, resolveKeys, resolveLipsyncOptions, resolveSpeechOptions, resolveTranscriptionOptions, resolveVideoOptions, resolveVoiceChangeOptions, resolveVoiceCloneOptions, resolveVoiceDesignOptions, shouldRetry, sleep, sortCatalogRows, streamObject, streamText, takesTargetAspectRatio, textUsageEvent, toGeneratedImage, validateInputSpec, videoLengthViolation, videoToolConstraints, withRetry, xaiVideoAdapter };