vertex_ai_gemini_routingTier 1 · 70% confidence

ai-agents-vertex-ai-gemini-rou-when-using-vertex-ai-gemini-models-e-g-gemini-2-5--fa6dbc00

agent: ai_agents

When does this happen?

IF When using Vertex AI Gemini models (e.g., gemini-2.5-flash-image-preview) with the /v1/images/generations endpoint, LiteLLM incorrectly routes to the :predict endpoint, causing a 404 error.

How others solved it

THEN Modify the image generation handler to detect Gemini models. Convert the request to a chat completion with modalities: ['text', 'image'] and use the :generateContent endpoint internally. After receiving the response, convert it back to the standard image generation format.

if model.startswith('vertex_ai/gemini'):
    # Convert to chat completion request
    chat_request = {
        'model': model,
        'messages': [{'role': 'user', 'content': prompt}],
        'modalities': ['text', 'image']
    }
    # Route to :generateContent endpoint
    response = await client.chat.completions.create(**chat_request)
    # Convert response to standard image generation format
    image_data = extract_image_from_response(response)
    return {'data': [{'url': image_data}]}

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics