Image model (gemini-3.1-flash-image) returns empty content in Anthropic format

[stephenpham68]

Description

When using gemini-3.1-flash-image via the Anthropic /v1/messages endpoint, the model generates an image successfully (output_tokens > 0) but the response content field is null or empty.

Steps to Reproduce

curl -X POST http://localhost:8317/api/provider/agy/v1/messages \
  -H 'Content-Type: application/json' \
  -H 'x-api-key: <key>' \
  -d '{
    "model": "gemini-3.1-flash-image",
    "max_tokens": 4096,
    "messages": [{"role": "user", "content": "Draw a red circle on white background."}]
  }'

Actual Result

{
  "content": null,
  "usage": {"input_tokens": 14, "output_tokens": 1376}
}

• Non-streaming: content: null despite output_tokens: 1376
• Streaming: message_start sent with token count, but no content_block_start/content_block_delta events follow

Expected Result

The generated image should be returned as a base64-encoded content block, e.g.:

{
  "content": [
    {
      "type": "image",
      "source": {
        "type": "base64",
        "media_type": "image/png",
        "data": "..."
      }
    }
  ]
}

Analysis

The Gemini API returns the image data successfully (tokens are consumed), but it appears the Gemini-to-Anthropic response translator drops inline image blocks since the standard Anthropic message format only defines text and tool_use content block types.

Environment

• CLIProxy version: v6.8.53-0
• Provider: Antigravity (agy)
• Model: gemini-3.1-flash-image
• Account tier: G1-ULTRA-TIER (model is registered and available in quota)