Bug type

Crash (process/app exits or hangs)

Beta release blocker

No

Summary

The image_generate tool returns a fetch failed error when calling the Google Gemini image generation model, but direct API calls to Google work successfully.

Error Message

All image generation models failed (2): 
  google/gemini-3.1-flash-image-preview: fetch failed 
  google/gemini-3-pro-image-preview: fetch failed

Steps to reproduce

response = requests.post( "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash-image-preview:generateContent?key=API_KEY", headers={"Content-Type": "application/json"}, json={ "contents": [{"parts": [{"text": "Draw a futuristic AI office"}]}], "generationConfig": {"responseModalities": ["TEXT", "IMAGE"]} } )

Successfully generates and saves image</code></pre>

Expected behavior

The image_generate tool should successfully generate images just like direct API calls.

Actual behavior

Returns fetch failed error, unable to generate images.

OpenClaw version

2026.3.31 (213a704)

Operating system

Ubuntu

Install method

No response

Model

google/gemini-3.1-flash-image-preview

Provider / routing chain

openclaw ➡️google

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

Root Cause Analysis

Hypothesized Causes

The image_generate tool in OpenClaw may have one of the following issues:

1. Incorrect API Endpoint

• May be using the Imagen predict endpoint instead of Gemini's generateContent

1. Missing Required Parameters

• Gemini image generation requires "generationConfig": {"responseModalities": ["TEXT", "IMAGE"]}
• OpenClaw may not be passing this parameter

1. Incompatible Request Format

• Request body structure may not match the format expected by Gemini API

1. Model Name Resolution Issue

• google/gemini-3.1-flash-image-preview may not be correctly mapped to the API endpoint

Workaround

Use the following alternative solution until the bug is fixed:

#!/usr/bin/env python3
import requests
import base64

def generate_image_gemini(prompt, api_key, output_path):
    endpoint = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash-image-preview:generateContent?key={api_key}"

    payload = {
        "contents": [{"parts": [{"text": prompt}]}],
        "generationConfig": {"responseModalities": ["TEXT", "IMAGE"]}
    }

    response = requests.post(endpoint, headers={"Content-Type": "application/json"}, json=payload)
    data = response.json()

    for part in data["candidates"][0]["content"]["parts"]:
        if "inlineData" in part and part["inlineData"]["mimeType"].startswith("image/"):
            image_data = base64.b64decode(part["inlineData"]["data"])
            with open(output_path, "wb") as f:
                f.write(image_data)
            return output_path

    raise Exception("No image in response")

Suggested Fixes

1. Verify image_generate Tool Implementation Logic

• Check if the generateContent endpoint is being used correctly
• Check if the responseModalities parameter is being passed

1. Update Model Mapping

• Ensure google/gemini-3.1-flash-image-preview is correctly mapped to the Gemini API

1. Add Detailed Error Logging

• Current fetch failed is too generic; should return specific HTTP error codes and response content

1. Test Coverage

• Add integration tests for the Google Gemini image generation API