What it does

ComfyScript is a Python frontend for ComfyUI, the node-based Stable Diffusion interface. It translates ComfyUI’s JSON workflows into readable Python scripts via a transpiler, then runs them through a runtime that talks to a ComfyUI server. You can also write workflows directly in Python, generate them programmatically, or use ComfyUI nodes as a function library in other projects.

The interesting bit

The transpiler doesn’t just dump JSON into dicts—it reconstructs linear Python code with proper variable dependencies, so a tangled multi-pass workflow collapses into something you can git diff. The runtime then executes that script against a live ComfyUI server, returning PIL images or saving metadata back into generated images. It’s essentially treating a visual GUI as a compile target.

Key highlights

• Transpiler converts ComfyUI workflows to Python automatically; scripts embed as image metadata
• Runtime supports local or remote ComfyUI servers, with uv one-liner execution without installing Python
• Works as a library: import ComfyUI nodes as Python functions for ML research or custom node debugging
• Generates ComfyUI workflows from scripts for use in the web UI—useful for programmatically building huge graphs
• Jupyter Notebook and ipywidgets/Solara UI integrations included

Caveats

• Installation is involved: four different install paths depending on whether you have ComfyUI, want it as a package, or are using uv
• [default] extras are mandatory for common dependencies; omitting it installs nothing useful
• Type stubs exist but the README shows truncated docs; some node compatibility gaps likely

Verdict

Worth a look if you’re hitting ComfyUI’s GUI ceiling—batching, version control, or LLM-generated workflows. Skip it if you’re happy clicking nodes and don’t need loops or diffs.