openclaw - 💡(How to fix) Fix Google provider: image/vision fails with 400 via OpenAI-compatible endpoint

TL;DR

Use the native endpoint (/v1beta/models/{model}:generateContent) instead of the OpenAI-compatible endpoint (/v1beta/openai/chat/completions) for image requests with gemini-2.5-flash or gemini-2.5-pro models.

Guidance

• Identify the endpoint used by OpenClaw for image requests and verify if it's the OpenAI-compatible endpoint.
• Update the OpenClaw configuration to use the native endpoint for image requests with the specified models.
• Test the updated configuration using the provided curl examples to ensure the native endpoint works as expected.
• Consider adding model-specific endpoint handling in OpenClaw to accommodate different model requirements.

Example

# Updated curl example using the native endpoint
curl https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=KEY \
  -d '{"contents":[{"parts":[{"text":"Describe"},{"inlineData":{"mimeType":"image/jpeg","data":"..."}}]}]}'

Notes

This solution assumes that the native endpoint supports all required features and functionality for the specified models. Additional testing and verification may be necessary to ensure compatibility.

Recommendation

Apply workaround: Use the native endpoint for image requests with gemini-2.5-flash and gemini-2.5-pro models, as it has been shown to work correctly in the provided examples.