Try the Flux Fusion AI Tool! https://tensor.art/template/783749162905840606
Basic comfy workflow https://tensor.art/workflows/782600611341315516
It is recommended to use simple/beta scheduler for low steps.
Improved from V1 with sharper good quality images from 4 steps. This version merges Schnell + Finetuned Dev + Hyper using the same but refined formula of variable block ratios from V1.
AIO (All in one) versions include UNET + VAE + CLIP L + T5XXL (fp8). Also known as Checkpoint or Compact version.
Using BNB NF4 & GGUF quants in ComfyUI requires installing custom nodes that add special model loaders:
UNET GGUF: https://github.com/city96/ComfyUI-GGUF
UNET NF4: https://github.com/DenkingOfficial/ComfyUI_UNet_bitsandbytes_NF4
AIO NF4: https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4
For using UNET versions, you also need to have the TEXT ENCODERS and VAE.
If you don't have them, download them from here:
TEXT ENCODERS: https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main
VAE: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main/vae
Place the model in "models/unet", both text encoders in "models/clip" and vae in "models/vae" folder.
In ComfyUI, use the standard flux workflow or add 'Load Diffusion Model', 'DualClipLoader' and 'Load VAE' nodes to replace the checkpoint loader and complete the setup.
In Forge, set the option "Diffusion in low bits" to "bnb-nf4"
