@ai-sdk/prodia 2.0.0-beta.3 → 2.0.0-beta.31

package/src/prodia-provider.ts

@@ -1,5 +1,7 @@
 import {
+  type Experimental_VideoModelV4,
   type ImageModelV4,
+  type LanguageModelV4,
   NoSuchModelError,
   type ProviderV4,
 } from '@ai-sdk/provider';
@@ -11,6 +13,10 @@ import {
 } from '@ai-sdk/provider-utils';
 import { ProdiaImageModel } from './prodia-image-model';
 import type { ProdiaImageModelId } from './prodia-image-settings';
+import { ProdiaLanguageModel } from './prodia-language-model';
+import type { ProdiaLanguageModelId } from './prodia-language-model-settings';
+import { ProdiaVideoModel } from './prodia-video-model';
+import type { ProdiaVideoModelId } from './prodia-video-model-settings';
 import { VERSION } from './version';
 
 export interface ProdiaProviderSettings {
@@ -37,6 +43,11 @@ export interface ProdiaProviderSettings {
 }
 
 export interface ProdiaProvider extends ProviderV4 {
+  /**
+   * Creates a language model for multimodal generation (img2img with text+image output).
+   */
+  languageModel(modelId: ProdiaLanguageModelId): LanguageModelV4;
+
   /**
    * Creates a model for image generation.
    */
@@ -47,6 +58,16 @@ export interface ProdiaProvider extends ProviderV4 {
    */
   imageModel(modelId: ProdiaImageModelId): ImageModelV4;
 
+  /**
+   * Creates a model for video generation.
+   */
+  video(modelId: ProdiaVideoModelId): Experimental_VideoModelV4;
+
+  /**
+   * Creates a model for video generation.
+   */
+  videoModel(modelId: ProdiaVideoModelId): Experimental_VideoModelV4;
+
   /**
    * @deprecated Use `embeddingModel` instead.
    */
@@ -80,25 +101,36 @@ export function createProdia(
       fetch: options.fetch,
     });
 
-  const embeddingModel = (modelId: string) => {
-    throw new NoSuchModelError({
-      modelId,
-      modelType: 'embeddingModel',
+  const createLanguageModel = (modelId: ProdiaLanguageModelId) =>
+    new ProdiaLanguageModel(modelId, {
+      provider: 'prodia.language',
+      baseURL: baseURL ?? defaultBaseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+    });
+
+  const createVideoModel = (modelId: ProdiaVideoModelId) =>
+    new ProdiaVideoModel(modelId, {
+      provider: 'prodia.video',
+      baseURL: baseURL ?? defaultBaseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
     });
-  };
 
-  const languageModel = (modelId: string) => {
+  const embeddingModel = (modelId: string) => {
     throw new NoSuchModelError({
       modelId,
-      modelType: 'languageModel',
+      modelType: 'embeddingModel',
     });
   };
 
   return {
     specificationVersion: 'v4',
+    languageModel: createLanguageModel,
     imageModel: createImageModel,
     image: createImageModel,
-    languageModel,
+    videoModel: createVideoModel,
+    video: createVideoModel,
     embeddingModel,
     textEmbeddingModel: embeddingModel,
   };

package/src/prodia-video-model-settings.ts

@@ -0,0 +1,7 @@
+/**
+ * Prodia job types for video generation.
+ */
+export type ProdiaVideoModelId =
+  | 'inference.wan2-2.lightning.txt2vid.v0'
+  | 'inference.wan2-2.lightning.img2vid.v0'
+  | (string & {});

package/src/prodia-video-model.ts

@@ -0,0 +1,282 @@
+import type {
+  Experimental_VideoModelV4,
+  SharedV4Warning,
+} from '@ai-sdk/provider';
+import type { InferSchema } from '@ai-sdk/provider-utils';
+import type { FetchFunction } from '@ai-sdk/provider-utils';
+import {
+  combineHeaders,
+  convertBase64ToUint8Array,
+  lazySchema,
+  parseJSON,
+  parseProviderOptions,
+  postFormDataToApi,
+  postToApi,
+  resolve,
+  zodSchema,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+import type { ProdiaModelConfig } from './prodia-api';
+import {
+  buildProdiaProviderMetadata,
+  parseMultipart,
+  prodiaFailedResponseHandler,
+  prodiaJobResultSchema,
+} from './prodia-api';
+import type { ProdiaJobResult } from './prodia-api';
+import type { ProdiaVideoModelId } from './prodia-video-model-settings';
+
+export class ProdiaVideoModel implements Experimental_VideoModelV4 {
+  readonly specificationVersion = 'v4';
+  readonly maxVideosPerCall = 1;
+
+  get provider(): string {
+    return this.config.provider;
+  }
+
+  constructor(
+    readonly modelId: ProdiaVideoModelId,
+    private readonly config: ProdiaModelConfig,
+  ) {}
+
+  async doGenerate(
+    options: Parameters<Experimental_VideoModelV4['doGenerate']>[0],
+  ): Promise<Awaited<ReturnType<Experimental_VideoModelV4['doGenerate']>>> {
+    const warnings: Array<SharedV4Warning> = [];
+
+    const prodiaOptions = await parseProviderOptions({
+      provider: 'prodia',
+      providerOptions: options.providerOptions,
+      schema: prodiaVideoModelOptionsSchema,
+    });
+
+    const jobConfig: Record<string, unknown> = {};
+
+    if (options.prompt !== undefined) {
+      jobConfig.prompt = options.prompt;
+    }
+    if (options.seed !== undefined) {
+      jobConfig.seed = options.seed;
+    }
+    if (prodiaOptions?.resolution !== undefined) {
+      jobConfig.resolution = prodiaOptions.resolution;
+    }
+
+    const body = {
+      type: this.modelId,
+      config: jobConfig,
+    };
+
+    const currentDate = this.config._internal?.currentDate?.() ?? new Date();
+    const combinedHeaders = combineHeaders(
+      await resolve(this.config.headers),
+      options.headers,
+    );
+
+    let multipartResult: {
+      jobResult: ProdiaJobResult;
+      videoBytes: Uint8Array;
+      videoMediaType: string;
+    };
+    let responseHeaders: Record<string, string> | undefined;
+
+    if (options.image) {
+      // img2vid: multipart form-data request
+      const imageData = await resolveVideoFileData(
+        options.image,
+        this.config.fetch,
+      );
+      const formData = new FormData();
+      formData.append(
+        'job',
+        new Blob([JSON.stringify(body)], { type: 'application/json' }),
+        'job.json',
+      );
+      formData.append(
+        'input',
+        new Blob([imageData.bytes], { type: imageData.mediaType }),
+        'input' + getExtension(imageData.mediaType),
+      );
+
+      const result = await postFormDataToApi({
+        url: `${this.config.baseURL}/job?price=true`,
+        headers: {
+          ...combinedHeaders,
+          Accept: 'multipart/form-data; video/mp4',
+        },
+        formData,
+        failedResponseHandler: prodiaFailedResponseHandler,
+        successfulResponseHandler: createVideoMultipartResponseHandler(),
+        abortSignal: options.abortSignal,
+        fetch: this.config.fetch,
+      });
+
+      multipartResult = result.value;
+      responseHeaders = result.responseHeaders;
+    } else {
+      // txt2vid: JSON request
+      const result = await postToApi({
+        url: `${this.config.baseURL}/job?price=true`,
+        headers: {
+          ...combinedHeaders,
+          Accept: 'multipart/form-data; video/mp4',
+          'Content-Type': 'application/json',
+        },
+        body: {
+          content: JSON.stringify(body),
+          values: body,
+        },
+        failedResponseHandler: prodiaFailedResponseHandler,
+        successfulResponseHandler: createVideoMultipartResponseHandler(),
+        abortSignal: options.abortSignal,
+        fetch: this.config.fetch,
+      });
+
+      multipartResult = result.value;
+      responseHeaders = result.responseHeaders;
+    }
+
+    const { jobResult, videoBytes, videoMediaType } = multipartResult;
+
+    return {
+      videos: [
+        {
+          type: 'binary',
+          data: videoBytes,
+          mediaType: videoMediaType,
+        },
+      ],
+      warnings,
+      providerMetadata: {
+        prodia: {
+          videos: [buildProdiaProviderMetadata(jobResult)],
+        },
+      },
+      response: {
+        modelId: this.modelId,
+        timestamp: currentDate,
+        headers: responseHeaders,
+      },
+    };
+  }
+}
+
+export const prodiaVideoModelOptionsSchema = lazySchema(() =>
+  zodSchema(
+    z.object({
+      /**
+       * Video resolution (e.g. "480p", "720p").
+       */
+      resolution: z.string().optional(),
+    }),
+  ),
+);
+
+export type ProdiaVideoModelOptions = InferSchema<
+  typeof prodiaVideoModelOptionsSchema
+>;
+
+interface VideoMultipartResult {
+  jobResult: ProdiaJobResult;
+  videoBytes: Uint8Array;
+  videoMediaType: string;
+}
+
+function createVideoMultipartResponseHandler() {
+  return async ({
+    response,
+  }: {
+    response: Response;
+  }): Promise<{
+    value: VideoMultipartResult;
+    responseHeaders: Record<string, string>;
+  }> => {
+    const contentType = response.headers.get('content-type') ?? '';
+    const responseHeaders: Record<string, string> = {};
+    response.headers.forEach((value, key) => {
+      responseHeaders[key] = value;
+    });
+
+    const boundaryMatch = contentType.match(/boundary=([^\s;]+)/);
+    if (!boundaryMatch) {
+      throw new Error(
+        `Prodia response missing multipart boundary in content-type: ${contentType}`,
+      );
+    }
+    const boundary = boundaryMatch[1];
+
+    const arrayBuffer = await response.arrayBuffer();
+    const bytes = new Uint8Array(arrayBuffer);
+
+    const parts = parseMultipart(bytes, boundary);
+
+    let jobResult: ProdiaJobResult | undefined;
+    let videoBytes: Uint8Array | undefined;
+    let videoMediaType = 'video/mp4';
+
+    for (const part of parts) {
+      const contentDisposition = part.headers['content-disposition'] ?? '';
+      const partContentType = part.headers['content-type'] ?? '';
+
+      if (contentDisposition.includes('name="job"')) {
+        const jsonStr = new TextDecoder().decode(part.body);
+        jobResult = await parseJSON({
+          text: jsonStr,
+          schema: zodSchema(prodiaJobResultSchema),
+        });
+      } else if (contentDisposition.includes('name="output"')) {
+        videoBytes = part.body;
+        if (partContentType.startsWith('video/')) {
+          videoMediaType = partContentType;
+        }
+      } else if (partContentType.startsWith('video/')) {
+        videoBytes = part.body;
+        videoMediaType = partContentType;
+      }
+    }
+
+    if (!jobResult) {
+      throw new Error('Prodia multipart response missing job part');
+    }
+    if (!videoBytes) {
+      throw new Error('Prodia multipart response missing output video');
+    }
+
+    return {
+      value: { jobResult, videoBytes, videoMediaType },
+      responseHeaders,
+    };
+  };
+}
+
+async function resolveVideoFileData(
+  file: NonNullable<
+    Parameters<Experimental_VideoModelV4['doGenerate']>[0]['image']
+  >,
+  fetchFunction?: FetchFunction,
+): Promise<{ bytes: Uint8Array; mediaType: string }> {
+  if (file.type === 'file') {
+    const data =
+      typeof file.data === 'string'
+        ? convertBase64ToUint8Array(file.data)
+        : file.data;
+    return { bytes: data, mediaType: file.mediaType };
+  }
+  // URL type - fetch the data
+  const response = await (fetchFunction ?? globalThis.fetch)(file.url);
+  const arrayBuffer = await response.arrayBuffer();
+  const mediaType =
+    response.headers.get('content-type') ?? 'application/octet-stream';
+  return { bytes: new Uint8Array(arrayBuffer), mediaType };
+}
+
+function getExtension(mediaType: string): string {
+  const map: Record<string, string> = {
+    'image/png': '.png',
+    'image/jpeg': '.jpg',
+    'image/webp': '.webp',
+    'video/mp4': '.mp4',
+    'video/webm': '.webm',
+  };
+  return map[mediaType] ?? '';
+}

package/dist/index.d.mts

@@ -1,58 +0,0 @@
-import * as _ai_sdk_provider_utils from '@ai-sdk/provider-utils';
-import { InferSchema, FetchFunction } from '@ai-sdk/provider-utils';
-import { ProviderV4, ImageModelV4 } from '@ai-sdk/provider';
-
-/**
- * Prodia job types for image generation.
- */
-type ProdiaImageModelId = 'inference.flux-fast.schnell.txt2img.v2' | 'inference.flux.schnell.txt2img.v2' | (string & {});
-
-declare const prodiaImageModelOptionsSchema: _ai_sdk_provider_utils.LazySchema<{
-    steps?: number | undefined;
-    width?: number | undefined;
-    height?: number | undefined;
-    stylePreset?: "3d-model" | "analog-film" | "anime" | "cinematic" | "comic-book" | "digital-art" | "enhance" | "fantasy-art" | "isometric" | "line-art" | "low-poly" | "neon-punk" | "origami" | "photographic" | "pixel-art" | "texture" | "craft-clay" | undefined;
-    loras?: string[] | undefined;
-    progressive?: boolean | undefined;
-}>;
-type ProdiaImageModelOptions = InferSchema<typeof prodiaImageModelOptionsSchema>;
-
-interface ProdiaProviderSettings {
-    /**
-     * Prodia API key. Default value is taken from the `PRODIA_TOKEN` environment variable.
-     */
-    apiKey?: string;
-    /**
-     * Base URL for the API calls. Defaults to `https://inference.prodia.com/v2`.
-     */
-    baseURL?: string;
-    /**
-     * Custom headers to include in the requests.
-     */
-    headers?: Record<string, string>;
-    /**
-     * Custom fetch implementation. You can use it as a middleware to intercept
-     * requests, or to provide a custom fetch implementation for e.g. testing.
-     */
-    fetch?: FetchFunction;
-}
-interface ProdiaProvider extends ProviderV4 {
-    /**
-     * Creates a model for image generation.
-     */
-    image(modelId: ProdiaImageModelId): ImageModelV4;
-    /**
-     * Creates a model for image generation.
-     */
-    imageModel(modelId: ProdiaImageModelId): ImageModelV4;
-    /**
-     * @deprecated Use `embeddingModel` instead.
-     */
-    textEmbeddingModel(modelId: string): never;
-}
-declare function createProdia(options?: ProdiaProviderSettings): ProdiaProvider;
-declare const prodia: ProdiaProvider;
-
-declare const VERSION: string;
-
-export { type ProdiaImageModelId, type ProdiaImageModelOptions, type ProdiaImageModelOptions as ProdiaImageProviderOptions, type ProdiaProvider, type ProdiaProviderSettings, VERSION, createProdia, prodia };