FLUX.2-klein-4B-INT8-Comfy

INT8 Quantized model of FLUX.2-klein-4B for ComfyUI

Generation speed

• Tested on
  • RTX5090 (400W), ComfyUI with torch2.10.0+cu130
  • RTX3090 (280W), ComfyUI with torch2.9.1+cu130
  • RTX3060 (PCIe4.0 x4), ComfyUI with torch2.9.1+cu130
• Generates 1216x1856, 4steps, cfg 1.0
• Second~Fifth run measured by varying only the seed

Result

<!-- T2I Prompt: A professional anime-style illustration features a young woman with long, straight blue hair and bright blue eyes, wearing a blue and white military-inspired uniform. The uniform includes a cropped blue jacket with gold trim, gold epaulets, gold buttons, and a blue skirt with white accents and gold trim. She has a white belt with gold details around her waist, white thigh-high stockings, and a blue and white hat adorned with a gold star and white wings. The subject is positioned centrally in the frame, standing upright with her left hand raised near her face, making a peace sign with her index and middle fingers, while her right hand hangs naturally by her side. The background is a soft, blurred outdoor scene with green trees and a light blue sky, creating a bright and airy atmosphere with natural daylight illuminating the subject evenly, highlighting the glossy texture of her uniform and the smoothness of her skin. The color palette is dominated by blue and white, accented with gold, creating a vibrant yet cohesive look. --> <!-- EDIT Prompt: turn the woman 40 degree. fill the background. -->

How to run

1. pull https://github.com/BobJohnson24/ComfyUI-Flux2-INT8 on ComfyUI/custom_nodes.
2. run ComfyUI and search INT8.
3. Use the Load Diffusion Model INT8 (W8A8) node for model loading.
4. Use the Load LoRA INT8 nodes for Lora loading.

How to reproduce

1. install https://github.com/silveroxides/convert_to_quant

pip install convert_to_quant

1. quantize using bf16 model.

convert_to_quant -i models/flux-2-klein-4b.safetensors --int8 --block_size 128 --comfy_quant --flux2 --scaling_mode tensor

1. model will be saved as models/flux-2-klein-4b_learned_int8mixed_tensorwise.safetensors