AsymFlow Models for Pixel-Space Generation

Workflows

This repo provides image generation workflows based on FLUX.2 klein Base 9B.

AsymFLUX.2 klein 9B

AsymFLUX.2 klein 9B is an adapter that converts FLUX.2 klein Base 9B into a pixel-space AsymFlow model for text-to-image generation.

We provide three variants of the adapter:

• AsymFLUX.2 klein 9B

  The base adapter, trained on a LAION subset for 15k steps (batch size=256).

  • The most raw, realistic and versatile model.
  • Results are highly diverse and creative.
  • Minimal aesthetic bias. Requires careful prompting to achieve certain styles.
  • Text rendering and anatomy (e.g., fingers) are not very good, since the original model (FLUX.2 klein Base 9B) is not good at these aspects.

  Please download the image below and drag it into ComfyUI to load the workflow.

  

• AsymFLUX.2 klein 9B SFT Z-Image Turbo

  Finetuned on synthetic data generated by Z-Image Turbo for 500 steps (batch size=192).

  • Results are less diverse.
  • Text rendering and anatomy (e.g., fingers) are more stable due to reduced diversity.
  • Styles are more consistent and less sensitive to prompt changes.

  Please download the image below and drag it into ComfyUI to load the workflow.

  

• AsymFLUX.2 klein 9B SFT FLUX.2 klein

  Finetuned on synthetic data generated by FLUX.2 klein Ditilled 9B for 500 steps (batch size=192).

  • Results are less diverse.
  • Text rendering and anatomy (e.g., fingers) are more stable due to reduced diversity.
  • Styles are more consistent and less sensitive to prompt changes.

  Please download the image below and drag it into ComfyUI to load the workflow.

  

Model links

Base model

• Download flux-2-klein-base-9b-fp8.safetensors and save it to
  models/diffusion_models/flux-2-klein-base-9b-fp8.safetensors

AsymFlow adapters

• Download AsymFLUX.2-klein-9B/diffusion_pytorch_model.safetensors and save it to
  models/loras/asymflux2_klein_9b.safetensors
• Download AsymFLUX.2-klein-9B-collection/asymflux2_klein_9b_sft_flux2_klein/diffusion_pytorch_model.safetensors and save it to
  asymflux2_klein_9b_sft_flux2_klein.safetensors
• Download AsymFLUX.2-klein-9B-collection/asymflux2_klein_9b_sft_zimage_turbo/diffusion_pytorch_model.safetensors and save it to
  asymflux2_klein_9b_sft_zimage_turbo.safetensors

Text encoder

• Download qwen_3_8b_fp8mixed.safetensors and save it to
  models/text_encoders/qwen_3_8b_fp8mixed.safetensors

GGUF Support

To load GGUF models, please install the custom nodes in ComfyUI-GGUF first.

Then, replace the Load AsymFlow Model node in the workflows with the Load AsymFlow Model (GGUF) node and select the corresponding GGUF model file.

Live pixel preview

AsymFlow workflows support precise previews of the generation process. Launch ComfyUI with the arguments --preview-method auto --preview-size 1024 to enable live pixel preview.