metadata
license: other
language:
- en
pipeline_tag: text-to-image
tags:
- flux
- flux.1
- flux.1-schnell
- flux.1-dev
- flux-merge
- merge
library_name: diffusers
Merge of Schnell and Dev variants of the Flux.1 model, almost all block are set to different ratios so it's not a perfect proportion of each model.
Recommended 4-8 steps.
Important
All in one versions include VAE + CLIP + T5XXL (fp8).
UNET versions (Model only) need Text Encoders and VAE, downloads:
- Text Encoders: https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main
- VAE: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main/vae
GGUF versions (Q4, Q5, Q8) require a custom node to work in comfy: https://github.com/city96/ComfyUI-GGUF
BNB versions (NF4, FP4) require a custom node to work in comfy:
- All in one versions: https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4
- UNET versions: https://github.com/DenkingOfficial/ComfyUI_UNet_bitsandbytes_NF4
Weights version comparison
- "Extreme closeup, frog face, star crystal structure, intricate designs, glowing hues. Extreme depth of field, celestial light, shimmering details, otherworldly charm, majestic elegance."
- Test parameters: 8 Steps, CFG 3.5, 1536x1536, seed 0
Samples schnell-fusion-dev (fp8-e4m3fn)
Samples comparing Fusion versions
