Skip to content

Load and run SDNQ quantized models in ComfyUI with 50-75% VRAM savings!

License

Notifications You must be signed in to change notification settings

EnragedAntelope/comfyui-sdnq

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

146 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ComfyUI-SDNQ

Load and run SDNQ quantized models in ComfyUI with 50-75% VRAM savings!

Run large models like FLUX.2, FLUX.1, SD3.5, Qwen-Image, and more on consumer hardware with significantly reduced VRAM requirements.

image

Features

  • All-in-one node - Select model, enter prompt, generate
  • 20+ pre-configured models with auto-download from HuggingFace
  • 50-75% VRAM savings with SDNQ quantization
  • Memory modes: GPU (fastest), balanced (12-16GB), lowvram (8GB)
  • LoRA support, image editing, 14 schedulers
  • Performance options: Triton acceleration, xFormers, VAE tiling

Installation

ComfyUI Manager (Recommended)

Search for "comfyui-sdnq" → Install → Restart ComfyUI

Manual

cd ComfyUI/custom_nodes/
git clone https://github.com/EnragedAntelope/comfyui-sdnq.git
cd comfyui-sdnq && pip install -r requirements.txt

Quick Start

  1. Add SDNQ Sampler node (under sampling/SDNQ)
  2. Select a model from dropdown (auto-downloads on first use)
  3. Enter your prompt → Queue Prompt → Done!

Hover over inputs for tooltips - all parameters are documented in the UI.

Models

30+ pre-quantized models available: FLUX.1, FLUX.2, Qwen-Image (including 2512 Dec update), Z-Image, GLM-Image, LTX-2 video, and more.

Browse all models: Disty0's SDNQ Collection

Video Models (Experimental)

LTX-2 video models are now supported. Set num_frames > 1 for video generation. Output is a batch of images (frames) that can be connected to video export nodes.

Performance

For best speed (30-80% faster), install Triton:

  • Linux: pip install triton
  • Windows: pip install triton-windows

Triton enables optimized quantized matmul operations. Enabled by default when available.

Scheduler tip: Use FlowMatchEulerDiscreteScheduler for FLUX/SD3/Qwen. Use DPMSolverMultistepScheduler for SDXL/SD1.5.

Troubleshooting

Model loading errors → Update libraries:

pip install --upgrade transformers diffusers

Newest models (FLUX.2-klein, GLM-Image, Qwen-Image-2512, LTX-2) → Build diffusers from source:

pip install git+https://github.com/huggingface/diffusers.git

This ensures you have the latest pipeline support for cutting-edge models.

Out of memory → Try balanced or lowvram memory mode, or use uint4 models.

Slow performance → Install Triton (see above), or try use_xformers=True.

Credits

SDNQ by Disty0 - All quantization technology is developed and maintained by Disty0.

About

Load and run SDNQ quantized models in ComfyUI with 50-75% VRAM savings!

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •  

Languages