Hi HF folks — I’m sharing Eye.Art Polyphemus as a workflow/API engineering project. It’s a hosted router, not an HF model or downloadable checkpoint: one conversational MCP entry point routes requests to image generation, reference-image edits, SVG creation, prompt ideas, and follow-up edits.
The main design problem is continuity. A follow-up like “change this image” needs the prior image and conversation context, while “make a new one” should start fresh. The API supports persistent sessions/conversations and asynchronous jobs: generation can return a job ID, and clients poll eye_art_image_status instead of accidentally submitting the same request again.
Connect and docs: MCP & API — Eye.Art · MCP endpoint: https://eye.art/api/eye-mcp No model weights are hosted on HF; these are hosted Eye.Art workflows behind a chat/API surface. I’d love feedback from people building Spaces or MCP clients: what interaction patterns work best for carrying reference images through edits and showing users long-running image jobs?