vercel/ai · error · Error

URL-based images are not supported for Amazon Bedrock image

Error message

URL-based images are not supported for Amazon Bedrock image editing. Please provide the image data directly.

What it means

For Bedrock image editing (e.g. Nova inpainting/outpainting/color-guided generation), the source image must be embedded in the request as base64 data. getBase64Data throws when given a file of type 'url' because Bedrock cannot fetch remote URLs itself. You must download the image and pass its bytes/data yourself.

Source

Thrown at packages/amazon-bedrock/src/amazon-bedrock-image-model.ts:306

          (response.status ? `Status: ${response.status}` : ''),
      );
    }

    return {
      images: response.images,
      warnings,
      response: {
        timestamp: currentDate,
        modelId: this.modelId,
        headers: responseHeaders,
      },
    };
  }
}

function getBase64Data(file: ImageModelV4File): string {
  if (file.type === 'url') {
    throw new Error(
      'URL-based images are not supported for Amazon Bedrock image editing. ' +
        'Please provide the image data directly.',
    );
  }

  if (file.data instanceof Uint8Array) {
    return convertUint8ArrayToBase64(file.data);
  }

  // Already base64 string
  return file.data;
}

// minimal version of the schema, focussed on what is needed for the implementation
// this approach limits breakages when the API changes and increases efficiency
const amazonBedrockImageResponseSchema = z.object({
  // Normal successful response
  images: z.array(z.string()).optional(),

View on GitHub (pinned to 69428b1f8b)

Solutions

1. Download the image yourself and pass it as data — fetch(url), then pass the Uint8Array/ArrayBuffer/base64 string in `images`
2. If the image lives in S3, retrieve it with the AWS SDK and pass the bytes
3. Keep URL support only for models that accept it; for Bedrock always provide raw image data

Example fix

// before
await generateImage({ model, prompt: 'make it snowy', images: [{ type: 'url', url: 'https://example.com/photo.png' }] });
// after
const res = await fetch('https://example.com/photo.png');
const data = new Uint8Array(await res.arrayBuffer());
await generateImage({ model, prompt: 'make it snowy', images: [{ type: 'data', data, mediaType: 'image/png' }] });

for (const img of images) {
  if (img.type === 'url') {
    throw new Error('Bedrock image editing requires image data, not URLs — download first');
  }
}

function isImageFileData(file: ImageModelV4File): boolean {
  return file.type === 'data';
}

When it happens

Trigger: Calling generateImage with a bedrock image model and an editing prompt while supplying the reference image via a URL — e.g. images: [{ type: 'url', url: 'https://...' }].

Common situations: Passing an S3/CDN/web image URL directly as the source image; assuming parity with providers that accept URLs server-side; migrating code from OpenAI image editing to Bedrock.

Related errors

• Unsupported task type: ${taskType}
• Amazon Bedrock request was moderated: ${reasons.join(', ')}
• Amazon Bedrock returned no images. Status: ${response.status
• AI_UnsupportedFunctionalityError
• Multiple system messages that are separated by user/assistan

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/9f2e3e8dbe058d37. Report an issue: GitHub.