Double check the exact model string you are using in the node. Gemini 2.0 Flash supports images, but some experimental or text-only variants do not. The model name must include vision support like gemini-2.0-flash-exp or similar.
Also verify your image is properly base64 encoded with correct MIME type prefix. If using Google AI node, try switching to HTTP Request node and manually construct the API call to confirm the model and payload format are correct.
Check Google AI Studio to see which exact models support multimodal input currently.