ComfyUI on Google Colab (Free Tier)

This repository provides a Google Colab notebook to run ComfyUI on the free Google Colab tier (under 15 GB VRAM).
The primary focus is to run Flux models efficiently within the Colab free tier limits.

---

Quick Start

Click the button below to open the notebook directly in Google Colab:

---

Key Features

• Runs ComfyUI seamlessly on Colab's free tier (Tesla T4 GPU).
• Pre-configured for Flux models (recommended FP8 version).
• Supports LoRAs by using smaller, quantized models.
• Tested with Q5 and Q4 GGUF quantized models from ComfyUI-GGUF.
• Works well with LoRAs and various pipelines.
• Access to additional quantized Flux models on Hugging Face.

---

Models

FP8 Model

• The FP8 version of Flux works without LoRA on free Colab.
• It’s recommended as the default model due to its balance of quality and performance.

Quantized Models (for LoRA support)

• Q5 and Q4 GGUF versions are tested and work well with LoRAs.
• More quantized models can be found at:
  https://huggingface.co/city96/FLUX.1-dev-gguf

---

Requirements

• Google Colab (Free Tier) with GPU runtime (Tesla T4 or similar).
• ~15 GB of VRAM (default free tier is sufficient).
• Internet access to download model files (automated in the notebook).

---

Usage

1. Click Open in Colab button above or open the notebook manually.
2. Run all cells to install dependencies and launch ComfyUI.
3. The notebook will start ComfyUI on a public URL using LocalTunnel.
4. Load your desired model (FP8, Q5, or Q4) and start generating images.

---

Links & References

• ComfyUI: https://github.com/comfyanonymous/ComfyUI
• Flux Model Guide: https://comfyanonymous.github.io/ComfyUI_examples/flux/
• ComfyUI-GGUF: https://github.com/city96/ComfyUI-GGUF
• Quantized Flux Models: https://huggingface.co/city96/FLUX.1-dev-gguf

---

License

This project follows the license of ComfyUI and related model repositories.

---