gemini-3.1-flash-image (GA) ignores imageConfig.imageSize on Vertex AI — always returns 1K

[EdamH]

Description

The GA model gemini-3.1-flash-image silently ignores imageConfig.imageSize when called via the Vertex AI endpoint. All requests return ~512×2064 output (1K resolution) regardless of
whether imageSize is set to "1K", "2K", or "4K".

The preview model (gemini-3.1-flash-image-preview) respects imageSize correctly using the identical request on the same project and endpoint.

This is similar to #1461 but affects the GA model, not the preview.

Environment

• SDK: @google/genai@2.8.0 (also reproduced with 1.39.0)
• Endpoint: Vertex AI (vertexai: true)
• Project location: global
• Node.js: v25.6.1
• OS: macOS Darwin 25.3.0

Reproduction

1. Via SDK (@google/genai)

import { GoogleGenAI } from '@google/genai';

const client = new GoogleGenAI({ vertexai: true, project: 'my-project', location: 'global' });

const response = await client.models.generateContent({
  model: 'gemini-3.1-flash-image',  // GA model
  contents: [{ role: 'user', parts: [{ text: 'Generate a tall narrow product image (1:4 ratio)' }] }],
  config: {
    responseModalities: ['TEXT', 'IMAGE'],
    imageConfig: { imageSize: '4K' },
  },
});
// Output: always ~512×2064 regardless of imageSize

2. Via raw REST (bypassing SDK entirely)

curl -X POST \
  "https://aiplatform.googleapis.com/v1/projects/MY_PROJECT/locations/global/publishers/google/models/gemini-3.1-flash-image:generateContent" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"role": "user", "parts": [{"text": "Generate a tall product image"}]}],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "imageConfig": {"imageSize": "4K"}
    }
  }'
// Output: always ~512×2064

Test Results

GA Model (gemini-3.1-flash-image) — imageSize Ignored

imageSize	Output Width	Output Height
1K	512	2064
2K	512	2064
4K	512	2064

Preview Model (gemini-3.1-flash-image-preview) — imageSize Respected

imageSize	Output Width	Output Height
1K	512	2064
2K	1024	4128
4K	2048	8256

Additional Findings

• Reproduced with both @google/genai@1.39.0 and @2.8.0 — same results.
• Reproduced via raw HTTP (no SDK) — confirms the issue is server-side, not SDK serialization.
• responseFormat.image.imageSize (from Gemini Developer API docs) returns HTTP 400 on Vertex AI: "Unknown name "responseFormat" at 'generation_config': Cannot find field" — so this
  format is not supported on Vertex.
• The preview model sunset is June 25, 2026, making this a blocker for migration to GA.

Expected Behavior

imageConfig.imageSize should control output resolution for the GA model, matching the preview model's behavior.

---

[Varun-S10]

Hi @EdamH, Thank you for reporting the issue and sharing your code. We have attempted to reproduce the issue using your provided code, but it is working as expected on our end. We tested both the gemini-3.1-flash-image and gemini-3.1-flash-image-preview models. Could you please confirm that you are using the latest version? If the issue persists after retrying the image generation, please provide additional information so we can investigate further.