project -> A Next.js whiteboard app where users draw on a canvas (tldraw), type a prompt, and click Enhance The canvas is exported as a base64 PNG, sent to an AI vision model to generate a detailed image prompt, which is then passed to Pollinations.ai to generate a refined image shown in a preview overlay. NEED ->We need a free vision API that accepts a base64 image + text prompt and returns a text response. OpenRouter keeps routing to wrong models. Looking for a reliable free visio...