//
Generate images with Black Forest Labs' Flux
Flux is the latest state-of-the-art image model from Black Forest Labs (the Stable Diffusion creators). It produces stunning results but requires more VRAM than SDXL.
ComfyUI has native Flux support.
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txtDownload from Hugging Face and place in models folder.
# Flux Schnell (fast, lower quality)
# Flux Dev (slower, better quality)
# Place in: ComfyUI/models/unet/
# Also need text encoders and VAE
# CLIP: models/clip/
# VAE: models/vae/đĄ Total download is ~25GB for Flux Dev with encoders.
Use the Flux workflow template in ComfyUI.
# Download flux workflow from:
# https://github.com/comfyanonymous/ComfyUI_examples
# Drag and drop the .json into ComfyUIEnter a prompt and generate!
Prompt: "a cyberpunk city at night with neon signs"
Steps: 20 (Schnell) or 50 (Dev)â CUDA out of memory
â Flux needs 12GB minimum. Use Schnell or enable CPU offloading.
â Generation is very slow
â Flux is compute-heavy. RTX 4090 generates in ~30s, 4070 Ti in ~60s.