Why I Built Rabbit-Hole (and Why You Might Like It)

#programming #opensource #comfyui #python

🐇 Down the Rabbit-Hole: A New Way to Run ComfyUI

Hi everyone 👋

I'm excited to introduce Rabbit-Hole, an open-source project designed to make ComfyUI much easier to integrate into real-world, production-like environments.

🎯 Why Rabbit-Hole?

ComfyUI is a fantastic tool for experimentation — modular, GPU-accelerated, and deeply customizable.

But when I tried to build an actual service or run batch workflows, I hit several walls:

Workflow graphs are hard to automate
ComfyUI server wasn’t designed for headless/batch processing
Scaling across EC2 or dockerized environments was painful
Node code is often hard to reuse or compose cleanly

So I built Rabbit-Hole — a new abstraction layer on top of ComfyUI’s core logic.
Instead of nodes, we define Tunnels, and instead of full workflows, we compose Executors.

⚙️ How It Works

Tunnels are modular functions (e.g. SDXL encoding, KSampler, VAE decode...)
Executors are pre-defined flows (e.g. T2I, IT2I + ControlNet, Upscaling with LoRA...)
All flows are Pythonic, testable, and work great in headless/batch/Dockerized setups.

You no longer need to send JSON workflows or spin up ComfyUI servers. Just call a class.

from rabbit_hole.executors import T2IExecutor

exe = T2IExecutor()
exe("a photo of a rabbit in a field", output_path="result.png")

🚀 What’s Next?

Rabbit-Hole is under active development.
Some recent additions:

✅ SDXL + IPAdapter Plus support
✅ ControlNet + Hint Image (Canny)
✅ Upscaling by Model
✅ LoRA composition

Upcoming:

🔄 RAG + LLM-powered image generation guidance
🐍 CLI / REST API interface
🧩 Tunnel Auto-generation from ComfyUI workflows

🌟 Try it Out!

GitHub: https://github.com/pupba/Rabbit-Hole

If you're building an AI image generation service, or want to take ComfyUI to the next level — give Rabbit-Hole a try.

I’d love your feedback, issues, and stars! ⭐
Thanks for reading — see you down the rabbit hole.

DEV Community