CanvasBot

CanvasBot is an autonomous, production-oriented design workflow that turns a natural-language brief into a print-oriented flyer, poster, or business card. It separates creative reasoning from deterministic rendering: Qwen produces validated creative decisions, Stable Diffusion creates raster artwork—including portraits requested in the brief—and Python places vector text into a bleed-sized PDF. The repository keeps the original notebooks as an experimentation history while exposing the validated approach through a versioned FastAPI service and a Streamlit client. The production interface submits asynchronous design jobs and reports the active workflow stage. The API defaults to one result per job to keep latency and GPU pressure predictable on the target 6 GB card. The output has vector typography, physical dimensions, safe zones, and bleed. It is not yet a press-certified PDF/X file: generated images are RGB and no printer-specific ICC profile or CMYK conversion is applied. A real print integration must add the target printer's color profile and preflight rules.