Intermittent "Cannot fetch content" for presigned S3 fileUri

[dani-mp]

Environment details

• Programming language: TypeScript (Node.js)
• OS: AWS Lambda (Amazon Linux)
• Language runtime version: Node.js 20.x
• Package version: @google/genai ^1.43.0

Steps to reproduce

1. Generate a presigned AWS S3 URL for a video object and call models.generateContent with model: "gemini-2.5-pro" and contents including:
   • { fileData: { fileUri: "<presigned-s3-url>" } }
   • { text: "<prompt>" }
2. Intermittently receive:

   {
     "error": {
       "code": 400,
       "message": "Cannot fetch content from the provided URL.",
       "status": "INVALID_ARGUMENT"
     }
   }

3. Immediately verify the same presigned URL is still valid and downloadable from our side.
4. Retry the exact same request; after automatic retries, it often succeeds with no URL/payload changes.

Observed behavior:

• The URL is a presigned AWS S3 URL (time-limited, signed query params).
• Failures are transient/non-deterministic for the same URL and prompt.
• We see this frequently in production; many failures resolve after retries.
• We want to confirm whether this is expected behavior for Gemini fetching presigned S3 URLs or a reliability issue with fileUri fetches.

[Varun-S10]

Hi @dani-mp, Thanks for bringing this issue to our attention. Could you please share a code snippet showing what you have tried so far so we can better understand the problem?

[dani-mp]

@Varun-S10

Here is a minimal version of what we are doing. presignedS3Url is a valid AWS S3 presigned URL for a video file.

import { GoogleGenAI } from '@google/genai'

const genAI = new GoogleGenAI({
  apiKey,
})

const response = await genAI.models.generateContent({
  model: 'gemini-2.5-pro',
  contents: [
    {
      fileData: {
        fileUri: presignedS3Url,
      },
    },
    {
      text: 'Analyze this video and return a short description.',
    },
  ],
  config: {
    responseMimeType: 'application/json',
  },
})

Intermittently, this returns:

{
  "error": {
    "code": 400,
    "message": "Cannot fetch content from the provided URL.",
    "status": "INVALID_ARGUMENT"
  }
}

The same presigned URL is still valid and downloadable from our side when this happens, and retrying the exact same request often succeeds.

[sanidhya-31]

Facing a very similar issue but in python

[dani-mp]

yeah, the issue is 100% an API issue. The API is quite unstable: apart from this error (it says 400, but it is not a bad request; it is a server-side issue), we get tons of 500s and 503s, and even some 429s, although we're far from hitting any limits within our tier.

[Varun-S10]

Hi @dani-mp, Thank you for bringing this to our attention. We have reproduced the issue. We truly appreciate you flagging this issue, and we will file a bug internally. In the meantime, you can use this simple workaround:

1. Download the video locally.
2. Upload it to the Gemini File API using genAI.files.upload().
3. Wait for Google to process the video.
4. Pass the new file.uri to generateContent().

Doing this bypasses the external link error so the model can successfully analyze your video.

[nagaozen]

2026-03-18T17:26:00.000Z bug has not been fixed yet.

[dani-mp]

This issue gets even more painful when working with batches. I have to partially retry a batch almost 10 times so I don't get this kind of error in some results.