Remesher: Automate Character Pipelines with ComfyUI's API, Docker, and Agents

Abstract

Remesher is a hands-on course about turning visual generative AI workflows into programmable character-production tools. Participants begin with a cloud-deployed ComfyUI environment, then build Remesher as a Python Typer CLI and agent-operable utility that submits workflows through ComfyUI’s API, monitors long-running jobs, and passes generated assets between image generation, image-to-3D reconstruction, texturing, and rigging stages. The course emphasizes a practical translation layer between artist-authored node graphs and production-oriented software: workflow JSON inspection, parameter overrides, artifact validation, failure recovery, and repeatable command-line execution. Building on prior SIGGRAPH work in ComfyUI education, cloud GPU deployment, and generative 3D integration, Remesher shifts attention from running visual workflows to packaging them as inspectable systems that humans and coding agents can operate. A central lesson is the path from CLI to agent skill: once a pipeline is exposed through clear commands, typed options, and observable outputs, agents can be taught to run, debug, and extend it without hiding the underlying workflow. Attendees leave with Remesher, a hands-on CLI and agent-operable utility, a reusable deployment pattern, and a model for making rapidly changing generative AI pipelines easier to reproduce, debug, and extend.