From e71b7f0eb9b0eff97037ef27542082a042fcc91b Mon Sep 17 00:00:00 2001 From: HM-RunningHub Date: Fri, 11 Apr 2025 18:18:19 +0800 Subject: [PATCH] Update README.md --- README.md | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 6ee2d8d..cb99ef7 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,7 @@ # UNO (Unity and Novel Output) for ComfyUI This repository hosts the ComfyUI implementation of UNO (Unity and Novel Output), supporting FLUX models. This implementation includes several new features and optimizations. +That can run the full version with 24GB VRAM, as well as quickly run the FP8 version. ## Features @@ -8,8 +9,8 @@ This repository hosts the ComfyUI implementation of UNO (Unity and Novel Output) - Support for `flux-dev-fp8` and `flux-schnell-fp8` - Note: `flux-schnell-fp8` offers lower consistency but much faster generation (4 steps) - **Memory optimization** through block swapping - - Run BF16 models on 24GB GPUs (actual usage: 14-20GB) - - Support for both `flux-dev` and `flux-schnell` in BF16 mode + - Run BF16 models on 24GB GPUs + - Support for both `flux-dev` and `flux-schnell` in BF16 mode on 24GB GPUs - **Progress bar** to display denoising progress in real-time - **Local model loading** configured via `config.json` @@ -18,7 +19,22 @@ This repository hosts the ComfyUI implementation of UNO (Unity and Novel Output) ### Directory Structure Models are configured in the root `config.json` file. The default structure expected is: - +``` +ComfyUI/models + flux/ + FLUX.1-schnell/ + text_encoder/ + tokenizer/ + text_encoder_2/ + tokenizer_2/ + unet/ + flux1-schnell.sft + flux1-dev.sft + vae/ + ae.safetensors + UNO/ + dit_lora.safetensors +``` ### Text Encoder and CLIP Configuration