@ai-sdk/prodia 2.0.0-beta.5 → 2.0.0-beta.54

package/src/prodia-language-model.ts

@@ -1,31 +1,37 @@
-import type {
-  LanguageModelV4,
-  LanguageModelV4CallOptions,
-  LanguageModelV4Content,
-  LanguageModelV4StreamPart,
-  SharedV4Warning,
+import {
+  UnsupportedFunctionalityError,
+  type LanguageModelV4,
+  type LanguageModelV4CallOptions,
+  type LanguageModelV4Content,
+  type LanguageModelV4StreamPart,
+  type SharedV4Warning,
 } from '@ai-sdk/provider';
-import type { InferSchema } from '@ai-sdk/provider-utils';
 import {
   combineHeaders,
+  isCustomReasoning,
   convertBase64ToUint8Array,
+  detectMediaType,
   generateId,
-  lazySchema,
+  getTopLevelMediaType,
+  isFullMediaType,
   parseJSON,
   parseProviderOptions,
   postFormDataToApi,
   resolve,
+  serializeModelOptions,
+  WORKFLOW_SERIALIZE,
+  WORKFLOW_DESERIALIZE,
   zodSchema,
 } from '@ai-sdk/provider-utils';
-import { z } from 'zod/v4';
-import type { ProdiaModelConfig } from './prodia-api';
 import {
   buildProdiaProviderMetadata,
   parseMultipart,
   prodiaFailedResponseHandler,
   prodiaJobResultSchema,
+  type ProdiaJobResult,
+  type ProdiaModelConfig,
 } from './prodia-api';
-import type { ProdiaJobResult } from './prodia-api';
+import { prodiaLanguageModelOptionsSchema } from './prodia-language-model-options';
 import type { ProdiaLanguageModelId } from './prodia-language-model-settings';
 
 export class ProdiaLanguageModel implements LanguageModelV4 {
@@ -36,6 +42,20 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
     return this.config.provider;
   }
 
+  static [WORKFLOW_SERIALIZE](model: ProdiaLanguageModel) {
+    return serializeModelOptions({
+      modelId: model.modelId,
+      config: model.config,
+    });
+  }
+
+  static [WORKFLOW_DESERIALIZE](options: {
+    modelId: ProdiaLanguageModelId;
+    config: ProdiaModelConfig;
+  }) {
+    return new ProdiaLanguageModel(options.modelId, options.config);
+  }
+
   constructor(
     readonly modelId: ProdiaLanguageModelId,
     private readonly config: ProdiaModelConfig,
@@ -79,6 +99,14 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
       warnings.push({ type: 'unsupported', feature: 'responseFormat' });
     }
 
+    if (isCustomReasoning(options.reasoning)) {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'reasoning',
+        details: 'This provider does not support reasoning configuration.',
+      });
+    }
+
     const prodiaOptions = await parseProviderOptions({
       provider: 'prodia',
       providerOptions: options.providerOptions,
@@ -116,19 +144,48 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
       const message = options.prompt[i];
       if (message.role === 'user') {
         for (const part of message.content) {
-          if (part.type === 'file' && part.mediaType.startsWith('image/')) {
-            if (part.data instanceof Uint8Array) {
-              imageBytes = part.data;
-            } else if (typeof part.data === 'string') {
-              // base64 encoded
-              imageBytes = convertBase64ToUint8Array(part.data);
-            } else if (part.data instanceof URL) {
-              const fetchFn = this.config.fetch ?? globalThis.fetch;
-              const response = await fetchFn(part.data.toString());
-              const arrayBuffer = await response.arrayBuffer();
-              imageBytes = new Uint8Array(arrayBuffer);
+          if (
+            part.type === 'file' &&
+            getTopLevelMediaType(part.mediaType) === 'image'
+          ) {
+            switch (part.data.type) {
+              case 'reference': {
+                throw new UnsupportedFunctionalityError({
+                  functionality: 'file parts with provider references',
+                });
+              }
+              case 'text': {
+                throw new UnsupportedFunctionalityError({
+                  functionality: 'text file parts',
+                });
+              }
+              case 'data': {
+                if (part.data.data instanceof Uint8Array) {
+                  imageBytes = part.data.data;
+                } else {
+                  imageBytes = convertBase64ToUint8Array(part.data.data);
+                }
+                break;
+              }
+              case 'url': {
+                const fetchFn = this.config.fetch ?? globalThis.fetch;
+                const response = await fetchFn(part.data.url.toString());
+                const arrayBuffer = await response.arrayBuffer();
+                imageBytes = new Uint8Array(arrayBuffer);
+                break;
+              }
+            }
+            if (isFullMediaType(part.mediaType)) {
+              imageMediaType = part.mediaType;
+            } else if (imageBytes !== undefined) {
+              const detected = detectMediaType({
+                data: imageBytes,
+                topLevelType: getTopLevelMediaType(part.mediaType),
+              });
+              if (detected !== undefined) {
+                imageMediaType = detected;
+              }
             }
-            imageMediaType = part.mediaType;
             break;
           }
         }
@@ -152,7 +209,7 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
 
     const currentDate = this.config._internal?.currentDate?.() ?? new Date();
     const combinedHeaders = combineHeaders(
-      await resolve(this.config.headers),
+      this.config.headers ? await resolve(this.config.headers) : undefined,
       options.headers,
     );
 
@@ -164,7 +221,7 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
       'job.json',
     );
     if (imageBytes) {
-      const ext =
+      const fileExtension =
         imageMediaType === 'image/png'
           ? '.png'
           : imageMediaType === 'image/jpeg'
@@ -175,7 +232,7 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
       formData.append(
         'input',
         new Blob([imageBytes], { type: imageMediaType }),
-        'input' + ext,
+        'input' + fileExtension,
       );
     }
 
@@ -204,7 +261,7 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
       content.push({
         type: 'file',
         mediaType: file.mediaType,
-        data: file.data,
+        data: { type: 'data', data: file.data },
       });
     }
 
@@ -291,35 +348,6 @@ export class ProdiaLanguageModel implements LanguageModelV4 {
   }
 }
 
-export const prodiaLanguageModelOptionsSchema = lazySchema(() =>
-  zodSchema(
-    z.object({
-      /**
-       * Aspect ratio for the output image.
-       */
-      aspectRatio: z
-        .enum([
-          '1:1',
-          '2:3',
-          '3:2',
-          '4:5',
-          '5:4',
-          '4:7',
-          '7:4',
-          '9:16',
-          '16:9',
-          '9:21',
-          '21:9',
-        ])
-        .optional(),
-    }),
-  ),
-);
-
-export type ProdiaLanguageModelOptions = InferSchema<
-  typeof prodiaLanguageModelOptionsSchema
->;
-
 interface LanguageMultipartResult {
   jobResult: ProdiaJobResult;
   textContent: string | undefined;

package/src/prodia-provider.ts

@@ -1,15 +1,15 @@
 import {
+  NoSuchModelError,
   type Experimental_VideoModelV4,
   type ImageModelV4,
   type LanguageModelV4,
-  NoSuchModelError,
   type ProviderV4,
 } from '@ai-sdk/provider';
-import type { FetchFunction } from '@ai-sdk/provider-utils';
 import {
   loadApiKey,
   withoutTrailingSlash,
   withUserAgentSuffix,
+  type FetchFunction,
 } from '@ai-sdk/provider-utils';
 import { ProdiaImageModel } from './prodia-image-model';
 import type { ProdiaImageModelId } from './prodia-image-settings';

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

@@ -0,0 +1,21 @@
+import {
+  lazySchema,
+  zodSchema,
+  type InferSchema,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+
+export const prodiaVideoModelOptionsSchema = lazySchema(() =>
+  zodSchema(
+    z.object({
+      /**
+       * Video resolution (e.g. "480p", "720p").
+       */
+      resolution: z.string().optional(),
+    }),
+  ),
+);
+
+export type ProdiaVideoModelOptions = InferSchema<
+  typeof prodiaVideoModelOptionsSchema
+>;

package/src/prodia-video-model.ts

@@ -2,12 +2,10 @@ 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,
+  downloadBlob,
   parseJSON,
   parseProviderOptions,
   postFormDataToApi,
@@ -15,15 +13,15 @@ import {
   resolve,
   zodSchema,
 } from '@ai-sdk/provider-utils';
-import { z } from 'zod/v4';
-import type { ProdiaModelConfig } from './prodia-api';
 import {
   buildProdiaProviderMetadata,
   parseMultipart,
   prodiaFailedResponseHandler,
   prodiaJobResultSchema,
+  type ProdiaJobResult,
+  type ProdiaModelConfig,
 } from './prodia-api';
-import type { ProdiaJobResult } from './prodia-api';
+import { prodiaVideoModelOptionsSchema } from './prodia-video-model-options';
 import type { ProdiaVideoModelId } from './prodia-video-model-settings';
 
 export class ProdiaVideoModel implements Experimental_VideoModelV4 {
@@ -84,7 +82,7 @@ export class ProdiaVideoModel implements Experimental_VideoModelV4 {
       // img2vid: multipart form-data request
       const imageData = await resolveVideoFileData(
         options.image,
-        this.config.fetch,
+        options.abortSignal,
       );
       const formData = new FormData();
       formData.append(
@@ -161,21 +159,6 @@ export class ProdiaVideoModel implements Experimental_VideoModelV4 {
   }
 }
 
-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;
@@ -253,7 +236,7 @@ async function resolveVideoFileData(
   file: NonNullable<
     Parameters<Experimental_VideoModelV4['doGenerate']>[0]['image']
   >,
-  fetchFunction?: FetchFunction,
+  abortSignal?: AbortSignal,
 ): Promise<{ bytes: Uint8Array; mediaType: string }> {
   if (file.type === 'file') {
     const data =
@@ -262,11 +245,12 @@ async function resolveVideoFileData(
         : 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';
+  // URL type - download via downloadBlob so the user-supplied URL is routed
+  // through the SSRF guard (validateDownloadUrl) instead of being fetched
+  // directly, preventing requests to private/internal addresses.
+  const blob = await downloadBlob(file.url, { abortSignal });
+  const arrayBuffer = await blob.arrayBuffer();
+  const mediaType = blob.type || 'application/octet-stream';
   return { bytes: new Uint8Array(arrayBuffer), mediaType };
 }
 

package/dist/index.d.mts

@@ -1,90 +0,0 @@
-import * as _ai_sdk_provider_utils from '@ai-sdk/provider-utils';
-import { InferSchema, FetchFunction } from '@ai-sdk/provider-utils';
-import { ProviderV4, LanguageModelV4, ImageModelV4, Experimental_VideoModelV4 } 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>;
-
-/**
- * Prodia job types for language model (multimodal img2img).
- */
-type ProdiaLanguageModelId = 'inference.nano-banana.img2img.v2' | (string & {});
-
-declare const prodiaLanguageModelOptionsSchema: _ai_sdk_provider_utils.LazySchema<{
-    aspectRatio?: "1:1" | "2:3" | "3:2" | "4:5" | "5:4" | "4:7" | "7:4" | "9:16" | "16:9" | "9:21" | "21:9" | undefined;
-}>;
-type ProdiaLanguageModelOptions = InferSchema<typeof prodiaLanguageModelOptionsSchema>;
-
-/**
- * Prodia job types for video generation.
- */
-type ProdiaVideoModelId = 'inference.wan2-2.lightning.txt2vid.v0' | 'inference.wan2-2.lightning.img2vid.v0' | (string & {});
-
-declare const prodiaVideoModelOptionsSchema: _ai_sdk_provider_utils.LazySchema<{
-    resolution?: string | undefined;
-}>;
-type ProdiaVideoModelOptions = InferSchema<typeof prodiaVideoModelOptionsSchema>;
-
-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 language model for multimodal generation (img2img with text+image output).
-     */
-    languageModel(modelId: ProdiaLanguageModelId): LanguageModelV4;
-    /**
-     * Creates a model for image generation.
-     */
-    image(modelId: ProdiaImageModelId): ImageModelV4;
-    /**
-     * Creates a model for image generation.
-     */
-    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.
-     */
-    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 ProdiaLanguageModelId, type ProdiaLanguageModelOptions, type ProdiaProvider, type ProdiaProviderSettings, type ProdiaVideoModelId, type ProdiaVideoModelOptions, VERSION, createProdia, prodia };