Mapping the Comfyverse: Using AI to Document Over 1600 Nodes

Using an LLM integrated within a Salt workflow, we've documented all 200+ ComfyUI nodes, plus 1400 custom nodes. And we're just getting started.
Alex Duffy
Alex Duffy
April 1, 2024

It's no secret that one of the biggest pain points when it comes to working with node-based AI tools like ComfyUI is understanding what nodes are available and how to use them effectively. With hundreds of default options and thousands of custom nodes, it can be incredibly daunting to find the right building blocks for your workflow, let alone to get them properly set up and configured.

This isn't just frustrating for ComfyUI veterans looking to experiment quickly. It also creates a major barrier to entry for newcomers, making the world of AI development unnecessarily opaque and inaccessible.

At Salt, we're committed to simplifying and demystifying the process of building with AI. After experiencing our own obstacles with inconsistent (or nonexistent) documentation, we decided to take a crack at providing comprehensive, user-friendly documentation powered by advanced AI. 

Our first step: documenting not just the 200+ default ComfyUI nodes, but over 1,400 custom nodes supported on the Salt platform. With the help of an LLM inside a Salt workflow, we were able to rapidly catalog thousands of nodes, including clear explanations for each node's functionality and usage.

That said, this AI-first approach comes with some pros and cons – we've released this resource to empower builders right away, but we need your help identifying gaps and errors. We’d also like your feedback and ideas on how to make this even more powerful and accessible for the community.

We know that good documentation isn’t just about listing features; it's about enabling developers to utilize them effectively. By establishing a "single source of truth" for node capabilities that goes beyond technical specs to include human-readable descriptions, examples, and other resources, we can equip language models with the knowledge required to empower more and more people to build AI solutions. 

This project lays the foundation for future advancements we’re exploring, including an LLM-powered "flow pilot" that can understand the building blocks of node-based development and guide users to construct sophisticated custom workflows through simple conversation. We’re excited for a future where creating advanced AI tooling will be as easy as describing your goals.

Check out our initial node documentation effort, kick the tires, and let us know what you think! We hope this is a helpful first step towards making the experience of building workflows a more intuitive process.

📚 Salt AI Node Documentation

✍️ Share your feedback and ideas in our Discord