ComfyUI_UltraFlux

ComfyUI_UltraFlux

UltraFlux:Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios,try it in comfyUI

Update

• 改为comfyUI 的核心封装，避免diffuser、transformer的迭代导致错误。/change to core encapsulation of comfyUI to avoid errors caused by the iteration of diffuser and transformer.
• 4096x4096支持,须latent设置为2048x2048 /4096x4096 support，requires latent to be 2048x2048

1.Installation

In the ./ComfyUI/custom_nodes directory, run the following:

git clone https://github.com/smthemex/ComfyUI_UltraFlux

2.requirements

• 无需安装额外依赖（GGUF 路径需要 ComfyUI_Dif_GGUF 插件 或者city96的）。

3.Model

• gguf and safetensor and int8 fp8 smthem/UltraFlux-v1-gguf → ComfyUI/models/gguf
• transformer safetensors v1 or v1.1 → ComfyUI/models/diffusion_models
• vae v1 → ComfyUI/models/vae
• comfyUI normal T5 and clip-l → ComfyUI/models/clip
• lora, any turbo and style flux lora（comfy 原生 LoraLoader 直挂 MODEL）

├── ComfyUI/models/gguf # or diffusion_models
|     ├── UltraFlux-v1-1-Q8_0.gguf # or BF16
├── ComfyUI/models/diffusion_models # or gguf
|     ├── UltraFlux-v1-1-BF16.safetensors # or e4m3fn of int8 or fp8
├── ComfyUI/models/vae
|        ├── UltraFlux.safetensors
├── ComfyUI/models/clip
|        ├── t5xxl_fp8_e4m3fn.safetensors
|        ├── clip_l.safetensors
├── ComfyUI/models/loras
|        ├── any turbo lora
|        ├── any style lora

4.Example

• old

5.Citation

@misc{ye2025ultrafluxdatamodelcodesignhighquality,
      title={UltraFlux: Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios},
      author={Tian Ye and Song Fei and Lei Zhu},
      year={2025},
      eprint={2511.18050},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.18050},
}