@runpod/ai-sdk-provider 0.11.0 → 0.12.0

package/CHANGELOG.md

@@ -1,5 +1,37 @@
 # @runpod/ai-sdk-provider
 
+## 0.12.0
+
+### Minor Changes
+
+- dcc2cc5: Add support for speech generation with `resembleai/chatterbox-turbo` model:
+  - `speechModel()` and `speech()` methods for text-to-speech
+  - Voice cloning via URL (5-10 seconds of audio)
+  - 20 built-in voices
+
+### Patch Changes
+
+- ace58c2: Add comprehensive documentation for Pruna and Nano Banana Pro models, including all supported aspect ratios, resolutions, and output formats. Update examples to use standard AI SDK options where possible.
+
+## 0.11.1
+
+### Patch Changes
+
+- f6115ac: Fix Pruna and Nano Banana Pro model support for all aspect ratios:
+
+  Pruna models:
+  - Skip standard size/aspectRatio validation
+  - Support all t2i aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, custom
+  - Support all edit aspect ratios: match_input_image, 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3
+  - Support custom width/height for t2i (256-1440, must be multiple of 16)
+  - Support 1-5 images for edit
+
+  Nano Banana Pro model:
+  - Skip standard size/aspectRatio validation
+  - Support all aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, 9:21
+  - Support resolution: 1k, 2k, 4k
+  - Support output_format: jpeg, png, webp
+
 ## 0.11.0
 
 ### Minor Changes

package/README.md

@@ -224,24 +224,69 @@ writeFileSync('landscape.jpg', image.uint8Array);
 
 ### Model Capabilities
 
-| Model ID                               | Description                     | Supported Aspect Ratios               |
-| -------------------------------------- | ------------------------------- | ------------------------------------- |
-| `bytedance/seedream-3.0`               | Advanced text-to-image model    | 1:1, 4:3, 3:4                         |
-| `bytedance/seedream-4.0`               | Text-to-image (v4)              | 1:1 (supports 1024, 2048, 4096)       |
-| `bytedance/seedream-4.0-edit`          | Image editing (v4, multi-image) | 1:1 (supports 1024, 1536, 2048, 4096) |
-| `black-forest-labs/flux-1-schnell`     | Fast image generation (4 steps) | 1:1, 4:3, 3:4                         |
-| `black-forest-labs/flux-1-dev`         | High-quality image generation   | 1:1, 4:3, 3:4                         |
-| `black-forest-labs/flux-1-kontext-dev` | Context-aware image generation  | 1:1, 4:3, 3:4                         |
-| `qwen/qwen-image`                      | Text-to-image generation        | 1:1, 4:3, 3:4                         |
-| `qwen/qwen-image-edit`                 | Image editing (prompt-guided)   | 1:1, 4:3, 3:4                         |
-| `nano-banana-edit`                     | Image editing (multi-image)     | 1:1, 4:3, 3:4                         |
-| `google/nano-banana-pro-edit`          | Image editing (Gemini-powered)  | Uses resolution param (1k, 2k)        |
-| `pruna/p-image-t2i`                    | Pruna text-to-image             | 1:1, 16:9, 9:16, 4:3, 3:4, etc.       |
-| `pruna/p-image-edit`                   | Pruna image editing             | match_input_image, 1:1, 16:9, etc.    |
-
-**Note**: The provider uses strict validation for image parameters. Unsupported aspect ratios (like `16:9`, `9:16`, `3:2`, `2:3`) will throw an `InvalidArgumentError` with a clear message about supported alternatives.
+| Model ID                               | Type |
+| -------------------------------------- | ---- |
+| `bytedance/seedream-3.0`               | t2i  |
+| `bytedance/seedream-4.0`               | t2i  |
+| `bytedance/seedream-4.0-edit`          | edit |
+| `black-forest-labs/flux-1-schnell`     | t2i  |
+| `black-forest-labs/flux-1-dev`         | t2i  |
+| `black-forest-labs/flux-1-kontext-dev` | edit |
+| `qwen/qwen-image`                      | t2i  |
+| `qwen/qwen-image-edit`                 | edit |
+| `nano-banana-edit`                     | edit |
+| `google/nano-banana-pro-edit`          | edit |
+| `pruna/p-image-t2i`                    | t2i  |
+| `pruna/p-image-edit`                   | edit |
+
+For the full list of models, see the [Runpod Public Endpoint Reference](https://docs.runpod.io/hub/public-endpoint-reference).
+
+### Pruna Models
+
+Supported models: `pruna/p-image-t2i`, `pruna/p-image-edit`
+
+| Parameter                                 | Supported Values                                  | Notes                                                 |
+| :---------------------------------------- | :------------------------------------------------ | :---------------------------------------------------- |
+| `aspectRatio`                             | `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `3:2`, `2:3` | Standard AI SDK parameter                             |
+| `aspectRatio` (t2i only)                  | `custom`                                          | Requires `width` & `height` in providerOptions        |
+| `providerOptions.runpod.width` / `height` | `256` - `1440`                                    | Custom dimensions (t2i only). Must be multiple of 16. |
+| `providerOptions.runpod.images`           | `string[]`                                        | Required for `p-image-edit`. Supports 1-5 images.     |
+
+**Example: Custom Resolution (t2i)**
 
-**Note:** This list is not complete. For a full list of all available models, see the [Runpod Public Endpoint Reference](https://docs.runpod.io/hub/public-endpoint-reference).
+```ts
+const { image } = await generateImage({
+  model: runpod.imageModel('pruna/p-image-t2i'),
+  prompt: 'A robot',
+  providerOptions: {
+    runpod: {
+      aspect_ratio: 'custom',
+      width: 512,
+      height: 768,
+    },
+  },
+});
+```
+
+### Google Models
+
+#### Nano Banana Pro
+
+Supported model: `google/nano-banana-pro-edit`
+
+| Parameter                       | Supported Values                                                  | Notes                             |
+| :------------------------------ | :---------------------------------------------------------------- | :-------------------------------- |
+| `aspectRatio`                   | `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `3:2`, `2:3`, `21:9`, `9:21` | Standard AI SDK parameter         |
+| `resolution`                    | `1k`, `2k`, `4k`                                                  | Output resolution quality         |
+| `output_format`                 | `jpeg`, `png`, `webp`                                             | Output image format               |
+| `providerOptions.runpod.images` | `string[]`                                                        | Required. Input image(s) to edit. |
+
+### Other Models
+
+Most other models (Flux, Seedream, Qwen, etc.) support standard `1:1`, `4:3`, and `3:4` aspect ratios.
+
+- **Flux models**: Support `num_inference_steps` and `guidance` settings.
+- **Edit models**: Require an input image via `providerOptions.runpod.image` (single) or `images` (multiple).
 
 ### Advanced Parameters
 
@@ -352,24 +397,122 @@ const { image } = await generateImage({
 
 ### Provider Options
 
-Runpod image models support flexible provider options through the `providerOptions.runpod` object:
-
-| Option                   | Type       | Default | Description                                                              |
-| ------------------------ | ---------- | ------- | ------------------------------------------------------------------------ |
-| `negative_prompt`        | `string`   | `""`    | Text describing what you don't want in the image                         |
-| `enable_safety_checker`  | `boolean`  | `true`  | Enable content safety filtering                                          |
-| `disable_safety_checker` | `boolean`  | `false` | Disable safety checker (Pruna models)                                    |
-| `image`                  | `string`   | -       | Single input image: URL or base64 data URI (Flux Kontext)                |
-| `images`                 | `string[]` | -       | Multiple input images (e.g., for `nano-banana-edit` multi-image editing) |
-| `aspect_ratio`           | `string`   | `"1:1"` | Aspect ratio string (Pruna: "16:9", "match_input_image", etc.)           |
-| `resolution`             | `string`   | `"1k"`  | Output resolution (Nano Banana Pro: "1k", "2k")                          |
-| `num_inference_steps`    | `number`   | Auto    | Number of denoising steps (Flux: 4 for schnell, 28 for others)           |
-| `guidance`               | `number`   | Auto    | Guidance scale for prompt adherence (Flux: 7 for schnell, 2 for others)  |
-| `output_format`          | `string`   | `"png"` | Output image format ("png", "jpg", or "jpeg")                            |
-| `enable_base64_output`   | `boolean`  | `false` | Return base64 instead of URL (Nano Banana Pro)                           |
-| `enable_sync_mode`       | `boolean`  | `false` | Enable synchronous mode (some models)                                    |
-| `maxPollAttempts`        | `number`   | `60`    | Maximum polling attempts for async generation                            |
-| `pollIntervalMillis`     | `number`   | `5000`  | Polling interval in milliseconds (5 seconds)                             |
+Use `providerOptions.runpod` for model-specific parameters:
+
+| Option                   | Type       | Default | Description                                     |
+| ------------------------ | ---------- | ------- | ----------------------------------------------- |
+| `negative_prompt`        | `string`   | `""`    | What to avoid in the image                      |
+| `enable_safety_checker`  | `boolean`  | `true`  | Content safety filtering                        |
+| `disable_safety_checker` | `boolean`  | `false` | Disable safety checker (Pruna)                  |
+| `image`                  | `string`   | -       | Input image URL or base64 (Flux Kontext)        |
+| `images`                 | `string[]` | -       | Multiple input images (edit models)             |
+| `resolution`             | `string`   | `"1k"`  | Output resolution: 1k, 2k, 4k (Nano Banana Pro) |
+| `width` / `height`       | `number`   | -       | Custom dimensions (Pruna t2i, 256-1440)         |
+| `num_inference_steps`    | `number`   | Auto    | Denoising steps                                 |
+| `guidance`               | `number`   | Auto    | Prompt adherence strength                       |
+| `output_format`          | `string`   | `"png"` | Output format: png, jpg, jpeg, webp             |
+| `maxPollAttempts`        | `number`   | `60`    | Max polling attempts                            |
+| `pollIntervalMillis`     | `number`   | `5000`  | Polling interval (ms)                           |
+
+## Speech
+
+You can generate speech using the AI SDK's `experimental_generateSpeech` and a Runpod speech model created via `runpod.speechModel()` (or the shorthand `runpod.speech()`).
+
+### Basic Usage
+
+```ts
+import { runpod } from '@runpod/ai-sdk-provider';
+import { experimental_generateSpeech as generateSpeech } from 'ai';
+
+const result = await generateSpeech({
+  model: runpod.speechModel('resembleai/chatterbox-turbo'),
+  text: 'Hello, this is Chatterbox Turbo running on Runpod.',
+  voice: 'lucy',
+});
+
+// Save to filesystem:
+import { writeFileSync } from 'fs';
+writeFileSync('speech.wav', result.audio.uint8Array);
+```
+
+**Returns:**
+
+- `result.audio.uint8Array` - Binary audio data (efficient for processing/saving)
+- `result.audio.base64` - Base64 encoded audio (useful for web embedding)
+- `result.audio.mediaType` - MIME type (e.g. `audio/wav`)
+- `result.audio.format` - Format (e.g. `wav`)
+- `result.warnings` - Array of any warnings about unsupported parameters
+- `result.providerMetadata.runpod.audioUrl` - Public URL to the generated audio
+- `result.providerMetadata.runpod.cost` - Cost information (if available)
+
+### Supported Models
+
+Supported model: `resembleai/chatterbox-turbo`
+
+### Parameters
+
+| Parameter | Type     | Default  | Description                              |
+| --------- | -------- | -------- | ---------------------------------------- |
+| `text`    | `string` | -        | Required. The text to convert to speech. |
+| `voice`   | `string` | `"lucy"` | Built-in voice name (see list below).    |
+
+### Provider Options
+
+Use `providerOptions.runpod` for model-specific parameters:
+
+| Option      | Type     | Default | Description                                 |
+| ----------- | -------- | ------- | ------------------------------------------- |
+| `voice_url` | `string` | -       | URL to audio file (5–10s) for voice cloning |
+| `voiceUrl`  | `string` | -       | Alias for `voice_url`                       |
+
+> Note: If `voice_url` is provided, the built-in `voice` is ignored.
+>
+> Note: This speech endpoint currently returns WAV only; `outputFormat` is ignored.
+
+### Voices
+
+`voice` selects one of the built-in voices (default: `lucy`):
+
+```ts
+[
+  'aaron',
+  'abigail',
+  'anaya',
+  'andy',
+  'archer',
+  'brian',
+  'chloe',
+  'dylan',
+  'emmanuel',
+  'ethan',
+  'evelyn',
+  'gavin',
+  'gordon',
+  'ivan',
+  'laura',
+  'lucy',
+  'madison',
+  'marisol',
+  'meera',
+  'walter',
+];
+```
+
+### Voice cloning (via URL)
+
+You can provide a `voice_url` (5–10s audio) through `providerOptions.runpod`:
+
+```ts
+const result = await generateSpeech({
+  model: runpod.speech('resembleai/chatterbox-turbo'),
+  text: 'Hello!',
+  providerOptions: {
+    runpod: {
+      voice_url: 'https://example.com/voice.wav',
+    },
+  },
+});
+```
 
 ## About Runpod
 

package/dist/index.d.mts

@@ -1,4 +1,4 @@
-import { LanguageModelV2, ImageModelV2 } from '@ai-sdk/provider';
+import { LanguageModelV2, ImageModelV2, SpeechModelV2 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
 export { OpenAICompatibleErrorData as RunpodErrorData } from '@ai-sdk/openai-compatible';
 import { z } from 'zod';
@@ -44,6 +44,14 @@ interface RunpodProvider {
   Creates an image model for image generation.
   */
     imageModel(modelId: string): ImageModelV2;
+    /**
+  Creates a speech model for speech generation.
+  */
+    speechModel(modelId: string): SpeechModelV2;
+    /**
+  Creates a speech model for speech generation.
+  */
+    speech(modelId: string): SpeechModelV2;
 }
 declare function createRunpod(options?: RunpodProviderSettings): RunpodProvider;
 declare const runpod: RunpodProvider;

package/dist/index.d.ts

@@ -1,4 +1,4 @@
-import { LanguageModelV2, ImageModelV2 } from '@ai-sdk/provider';
+import { LanguageModelV2, ImageModelV2, SpeechModelV2 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
 export { OpenAICompatibleErrorData as RunpodErrorData } from '@ai-sdk/openai-compatible';
 import { z } from 'zod';
@@ -44,6 +44,14 @@ interface RunpodProvider {
   Creates an image model for image generation.
   */
     imageModel(modelId: string): ImageModelV2;
+    /**
+  Creates a speech model for speech generation.
+  */
+    speechModel(modelId: string): SpeechModelV2;
+    /**
+  Creates a speech model for speech generation.
+  */
+    speech(modelId: string): SpeechModelV2;
 }
 declare function createRunpod(options?: RunpodProviderSettings): RunpodProvider;
 declare const runpod: RunpodProvider;

package/dist/index.js

@@ -27,7 +27,7 @@ module.exports = __toCommonJS(index_exports);
 
 // src/runpod-provider.ts
 var import_openai_compatible = require("@ai-sdk/openai-compatible");
-var import_provider_utils3 = require("@ai-sdk/provider-utils");
+var import_provider_utils4 = require("@ai-sdk/provider-utils");
 
 // src/runpod-image-model.ts
 var import_provider_utils2 = require("@ai-sdk/provider-utils");
@@ -115,8 +115,12 @@ var RunpodImageModel = class {
     abortSignal
   }) {
     const warnings = [];
+    const isPrunaModel = this.modelId.includes("pruna") || this.modelId.includes("p-image");
+    const isNanoBananaProModel = this.modelId.includes("nano-banana-pro");
     let runpodSize;
-    if (size) {
+    if (isPrunaModel || isNanoBananaProModel) {
+      runpodSize = aspectRatio || "1:1";
+    } else if (size) {
       const runpodSizeCandidate = size.replace("x", "*");
       if (!SUPPORTED_SIZES.has(runpodSizeCandidate)) {
         throw new import_provider.InvalidArgumentError({
@@ -150,7 +154,8 @@ var RunpodImageModel = class {
       prompt,
       runpodSize,
       seed,
-      providerOptions.runpod
+      providerOptions.runpod,
+      aspectRatio
     );
     const { value: response, responseHeaders } = await (0, import_provider_utils2.postJsonToApi)({
       url: `${this.config.baseURL}/runsync`,
@@ -264,7 +269,7 @@ var RunpodImageModel = class {
       `Image generation timed out after ${maxAttempts} attempts (${maxAttempts * pollInterval / 1e3}s)`
     );
   }
-  buildInputPayload(prompt, runpodSize, seed, runpodOptions) {
+  buildInputPayload(prompt, runpodSize, seed, runpodOptions, aspectRatio) {
     const isFluxModel = this.modelId.includes("flux") || this.modelId.includes("black-forest-labs");
     if (isFluxModel) {
       const isKontext = this.modelId.includes("kontext");
@@ -300,50 +305,56 @@ var RunpodImageModel = class {
     if (isPrunaModel) {
       const isPrunaEdit = this.modelId.includes("edit");
       if (isPrunaEdit) {
-        return {
+        const editPayload = {
           prompt,
-          seed: seed ?? -1,
-          aspect_ratio: runpodOptions?.aspect_ratio ?? "match_input_image",
-          disable_safety_checker: runpodOptions?.disable_safety_checker ?? false,
-          enable_sync_mode: runpodOptions?.enable_sync_mode ?? false,
-          ...runpodOptions
+          aspect_ratio: runpodOptions?.aspect_ratio ?? aspectRatio ?? "1:1",
+          disable_safety_checker: runpodOptions?.disable_safety_checker ?? false
         };
+        if (seed !== void 0) {
+          editPayload.seed = seed;
+        } else if (runpodOptions?.seed !== void 0) {
+          editPayload.seed = runpodOptions.seed;
+        }
+        if (runpodOptions?.images) {
+          editPayload.images = runpodOptions.images;
+        }
+        return editPayload;
       } else {
-        const aspectRatioMap = {
-          "1328*1328": "1:1",
-          "1472*1140": "4:3",
-          "1140*1472": "3:4",
-          "512*512": "1:1",
-          "768*768": "1:1",
-          "1024*1024": "1:1",
-          "1536*1536": "1:1",
-          "2048*2048": "1:1",
-          "4096*4096": "1:1",
-          "512*768": "2:3",
-          "768*512": "3:2",
-          "1024*768": "4:3",
-          "768*1024": "3:4"
-        };
-        const aspectRatio = runpodOptions?.aspect_ratio ?? aspectRatioMap[runpodSize] ?? "1:1";
-        return {
+        const t2iPayload = {
           prompt,
-          seed: seed ?? 0,
-          aspect_ratio: aspectRatio,
-          enable_safety_checker: runpodOptions?.enable_safety_checker ?? true,
-          ...runpodOptions
+          aspect_ratio: runpodOptions?.aspect_ratio ?? aspectRatio ?? "1:1",
+          disable_safety_checker: runpodOptions?.disable_safety_checker ?? false
         };
+        if (seed !== void 0) {
+          t2iPayload.seed = seed;
+        } else if (runpodOptions?.seed !== void 0) {
+          t2iPayload.seed = runpodOptions.seed;
+        }
+        if (t2iPayload.aspect_ratio === "custom") {
+          if (runpodOptions?.width) {
+            t2iPayload.width = runpodOptions.width;
+          }
+          if (runpodOptions?.height) {
+            t2iPayload.height = runpodOptions.height;
+          }
+        }
+        return t2iPayload;
       }
     }
     const isNanaBananaProModel = this.modelId.includes("nano-banana-pro");
     if (isNanaBananaProModel) {
-      return {
+      const nanoBananaPayload = {
         prompt,
+        aspect_ratio: runpodOptions?.aspect_ratio ?? aspectRatio ?? "1:1",
         resolution: runpodOptions?.resolution ?? "1k",
         output_format: runpodOptions?.output_format ?? "jpeg",
         enable_base64_output: runpodOptions?.enable_base64_output ?? false,
-        enable_sync_mode: runpodOptions?.enable_sync_mode ?? false,
-        ...runpodOptions
+        enable_sync_mode: runpodOptions?.enable_sync_mode ?? false
       };
+      if (runpodOptions?.images) {
+        nanoBananaPayload.images = runpodOptions.images;
+      }
+      return nanoBananaPayload;
     }
     return {
       prompt,
@@ -381,6 +392,148 @@ var runpodImageStatusSchema = import_zod2.z.object({
   // Error message if FAILED
 });
 
+// src/runpod-speech-model.ts
+var import_provider_utils3 = require("@ai-sdk/provider-utils");
+function isRecord(value) {
+  return typeof value === "object" && value !== null;
+}
+function replaceNewlinesWithSpaces(value) {
+  return value.replace(/[\r\n]+/g, " ");
+}
+var RunpodSpeechModel = class {
+  constructor(modelId, config) {
+    this.modelId = modelId;
+    this.config = config;
+    this.specificationVersion = "v2";
+  }
+  get provider() {
+    return this.config.provider;
+  }
+  getRunpodRunSyncUrl() {
+    const baseURL = (0, import_provider_utils3.withoutTrailingSlash)(this.config.baseURL) ?? this.config.baseURL;
+    if (baseURL.endsWith("/run") || baseURL.endsWith("/runsync")) {
+      return baseURL;
+    }
+    return `${baseURL}/runsync`;
+  }
+  async doGenerate(options) {
+    const currentDate = this.config._internal?.currentDate?.() ?? /* @__PURE__ */ new Date();
+    const warnings = [];
+    const {
+      text,
+      voice,
+      outputFormat,
+      instructions,
+      speed,
+      language,
+      providerOptions,
+      abortSignal,
+      headers
+    } = options;
+    if (outputFormat != null && outputFormat !== "wav") {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "outputFormat",
+        details: `Unsupported outputFormat: ${outputFormat}. This endpoint returns 'wav'.`
+      });
+    }
+    if (instructions != null) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "instructions",
+        details: `Instructions are not supported by this speech endpoint.`
+      });
+    }
+    if (speed != null) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "speed",
+        details: `Speed is not supported by this speech endpoint.`
+      });
+    }
+    if (language != null) {
+      warnings.push({
+        type: "unsupported-setting",
+        setting: "language",
+        details: `Language selection is not supported by this speech endpoint.`
+      });
+    }
+    const runpodProviderOptions = isRecord(providerOptions) ? providerOptions.runpod : void 0;
+    const voiceUrl = isRecord(runpodProviderOptions) && (typeof runpodProviderOptions.voice_url === "string" || typeof runpodProviderOptions.voiceUrl === "string") ? runpodProviderOptions.voice_url ?? runpodProviderOptions.voiceUrl ?? void 0 : void 0;
+    const input = {
+      prompt: replaceNewlinesWithSpaces(text)
+    };
+    if (voiceUrl) {
+      input.voice_url = voiceUrl;
+    } else if (voice) {
+      input.voice = voice;
+    }
+    const requestBody = { input };
+    const url = this.getRunpodRunSyncUrl();
+    const fetchFn = this.config.fetch ?? fetch;
+    const requestHeaders = {
+      "Content-Type": "application/json",
+      ...this.config.headers()
+    };
+    if (headers) {
+      for (const [key, value] of Object.entries(headers)) {
+        if (value != null) {
+          requestHeaders[key] = value;
+        }
+      }
+    }
+    const response = await fetchFn(url, {
+      method: "POST",
+      headers: requestHeaders,
+      body: JSON.stringify(requestBody),
+      signal: abortSignal
+    });
+    const responseHeaders = Object.fromEntries(response.headers.entries());
+    const rawBodyText = await response.text();
+    let parsed = void 0;
+    try {
+      parsed = rawBodyText ? JSON.parse(rawBodyText) : void 0;
+    } catch {
+    }
+    if (!response.ok) {
+      const message = parsed && typeof parsed.error === "string" && parsed.error || rawBodyText || `HTTP ${response.status}`;
+      throw new Error(`Runpod speech request failed: ${message}`);
+    }
+    const output = parsed?.output ?? parsed;
+    const audioUrl = output?.audio_url;
+    if (typeof audioUrl !== "string" || audioUrl.length === 0) {
+      throw new Error("Runpod speech response did not include an audio_url.");
+    }
+    const audioResponse = await fetchFn(audioUrl, { signal: abortSignal });
+    if (!audioResponse.ok) {
+      throw new Error(
+        `Failed to download generated audio (${audioResponse.status}).`
+      );
+    }
+    const audio = new Uint8Array(await audioResponse.arrayBuffer());
+    const providerMetadata = {
+      runpod: {
+        audioUrl,
+        ...typeof output?.cost === "number" ? { cost: output.cost } : {}
+      }
+    };
+    return {
+      audio,
+      warnings,
+      request: {
+        body: JSON.stringify(requestBody)
+      },
+      response: {
+        timestamp: currentDate,
+        modelId: this.modelId,
+        headers: responseHeaders,
+        body: rawBodyText
+      },
+      providerMetadata
+    };
+  }
+};
+
 // src/runpod-provider.ts
 var MODEL_ID_TO_ENDPOINT_URL = {
   "qwen/qwen3-32b-awq": "https://api.runpod.ai/v2/qwen3-32b-awq/openai/v1",
@@ -408,6 +561,9 @@ var IMAGE_MODEL_ID_TO_ENDPOINT_URL = {
   "pruna/p-image-t2i": "https://api.runpod.ai/v2/p-image-t2i",
   "pruna/p-image-edit": "https://api.runpod.ai/v2/p-image-edit"
 };
+var SPEECH_MODEL_ID_TO_ENDPOINT_URL = {
+  "resembleai/chatterbox-turbo": "https://api.runpod.ai/v2/chatterbox-turbo/"
+};
 var MODEL_ID_TO_OPENAI_NAME = {
   "qwen/qwen3-32b-awq": "Qwen/Qwen3-32B-AWQ",
   "deepcogito/cogito-671b-v2.1-fp8": "deepcogito/cogito-671b-v2.1-FP8",
@@ -417,9 +573,26 @@ function deriveEndpointURL(modelId) {
   const normalized = modelId.replace(/\//g, "-");
   return `https://api.runpod.ai/v2/${normalized}/openai/v1`;
 }
+function parseRunpodConsoleEndpointId(modelIdOrUrl) {
+  if (!modelIdOrUrl.startsWith("http")) {
+    return null;
+  }
+  try {
+    const url = new URL(modelIdOrUrl);
+    if (url.hostname !== "console.runpod.io") {
+      return null;
+    }
+    const parts = url.pathname.split("/").filter(Boolean);
+    const idx = parts.lastIndexOf("endpoint");
+    const endpointId = idx !== -1 ? parts[idx + 1] : void 0;
+    return endpointId || null;
+  } catch {
+    return null;
+  }
+}
 function createRunpod(options = {}) {
   const getHeaders = () => ({
-    Authorization: `Bearer ${(0, import_provider_utils3.loadApiKey)({
+    Authorization: `Bearer ${(0, import_provider_utils4.loadApiKey)({
       apiKey: options.apiKey,
       environmentVariableName: "RUNPOD_API_KEY",
       description: "Runpod"
@@ -449,7 +622,7 @@ function createRunpod(options = {}) {
     }
     return {
       provider: `runpod.${modelType}`,
-      url: ({ path }) => `${(0, import_provider_utils3.withoutTrailingSlash)(baseURL)}${path}`,
+      url: ({ path }) => `${(0, import_provider_utils4.withoutTrailingSlash)(baseURL)}${path}`,
       headers: getHeaders,
       fetch: runpodFetch
     };
@@ -482,11 +655,25 @@ function createRunpod(options = {}) {
       fetch: options.fetch
     });
   };
+  const createSpeechModel = (modelId) => {
+    const endpointIdFromConsole = parseRunpodConsoleEndpointId(modelId);
+    const normalizedModelId = endpointIdFromConsole ?? modelId;
+    const mappedBaseURL = SPEECH_MODEL_ID_TO_ENDPOINT_URL[normalizedModelId];
+    const baseURL = mappedBaseURL ?? (normalizedModelId.startsWith("http") ? normalizedModelId : `https://api.runpod.ai/v2/${normalizedModelId}`);
+    return new RunpodSpeechModel(normalizedModelId, {
+      provider: "runpod.speech",
+      baseURL,
+      headers: getHeaders,
+      fetch: runpodFetch
+    });
+  };
   const provider = (modelId) => createChatModel(modelId);
   provider.completionModel = createCompletionModel;
   provider.languageModel = createChatModel;
   provider.chatModel = createChatModel;
   provider.imageModel = createImageModel;
+  provider.speechModel = createSpeechModel;
+  provider.speech = createSpeechModel;
   return provider;
 }
 var runpod = createRunpod();