Hugging Face's logo

WAS
/
was-node-suite-weights

Diffusion Single File
comfyui
image-editing
image-restoration
hdr
denoising
low-light
optical-flow
video-generation
kandinsky
Model card Files
xet
 Community

WAS Node Suite weights

The model files WAS Node Suite runs in ComfyUI, kept in one place so a node does not break when a third-party copy is moved or removed. Some are upstream checkpoints byte for byte, some are converted to safetensors, and the model families under models/ are ComfyUI conversions and quantisations of upstream releases. Nothing here is trained by this project. Every file keeps the licence of the work it came from, listed beside it below; the Kandinsky 6 audio VAE and the BluePixel transformers are for non-commercial use only.

Folder What it holds Used by
denoise/ SCUNet and NAFNet denoisers Power Preprocessor
low_light/ HVI-CIDNet, Retinexformer and DarkIR low-light enhancers Power Preprocessor
hdr/ HDRCNN, converted to safetensors HDR Reconstruct
birefnet/, ben2/ BiRefNet and BEN2 background removal Image Remove Background
optical_flow/ FlowSeek optical flow, converted to safetensors Video Motion Model Loader
models/kandinsky6/ Kandinsky 6 video with sound: int8 ConvRot and W6A8 transformers, text encoders, audio VAE Load Diffusion Model and the Kandinsky 6 nodes
models/bluepixel/ BluePixel-Turbo 4K image editing: bf16, int8 ConvRot, W6A8 and W4A8 transformers, and its DC-AE autoencoder BluePixel Model Loader and BluePixel VAE Loader

With features.network: true in WAS Node Suite's config.yaml, the pack fetches a file from every folder outside models/ the first time a node needs it. The models/ folders are laid out as ComfyUI's own models folder: each subfolder goes into the matching folder under ComfyUI/models, downloaded by hand.

denoise/

File Original Licence
scunet_color_real_psnr.pth cszn/SCUNet, Kai Zhang Apache-2.0
NAFNet-SIDD-width32.pth megvii-research/NAFNet MIT, with Apache-2.0 for BasicSR
NAFNet-SIDD-width64.pth megvii-research/NAFNet MIT, with Apache-2.0 for BasicSR

hdr/

Converted to safetensors for native ComfyUI inference.

File Original Licence
hdrcnn.safetensors gabrieleilertsen/hdrcnn, Gabriel Eilertsen BSD-3-Clause

low_light/

File Original Licence
hvi-cidnet-generalization.safetensors Fediory/HVI-CIDNet MIT
hvi-cidnet-sice.safetensors Fediory/HVI-CIDNet MIT
hvi-cidnet-sony-total-dark.safetensors Fediory/HVI-CIDNet MIT
hvi-cidnet-fivek.safetensors Fediory/HVI-CIDNet MIT
retinexformer-ntire.pth caiyuanhao1998/Retinexformer MIT
retinexformer-lol-v1.pth same MIT
retinexformer-lol-v2-real.pth same MIT
retinexformer-lol-v2-synthetic.pth same MIT
retinexformer-fivek.pth same MIT
retinexformer-sid.pth same MIT
retinexformer-smid.pth same MIT
retinexformer-sdsd-indoor.pth same MIT
retinexformer-sdsd-outdoor.pth same MIT
darkir-m.pt cidautai/DarkIR MIT

birefnet/

File Original Licence
birefnet/General.safetensors ZhengPeng7/BiRefNet MIT
birefnet/General-HR.safetensors ZhengPeng7/BiRefNet MIT
birefnet/General-dynamic.safetensors ZhengPeng7/BiRefNet MIT
birefnet/General-reso_512.safetensors ZhengPeng7/BiRefNet MIT
birefnet/DIS.safetensors ZhengPeng7/BiRefNet MIT
birefnet/DIS-TR_TEs.safetensors ZhengPeng7/BiRefNet MIT
birefnet/COD.safetensors ZhengPeng7/BiRefNet MIT
birefnet/HRSOD.safetensors ZhengPeng7/BiRefNet MIT
birefnet/Matting-HR.safetensors ZhengPeng7/BiRefNet MIT
birefnet/Portrait.safetensors ZhengPeng7/BiRefNet MIT

ben2/

File Original Licence
ben2/ben2-base.safetensors PramaLLC/BEN2 MIT

optical_flow/

FlowSeek's released checkpoints, converted to safetensors with every tensor unchanged. Each carries the Depth Anything V2 Small weights FlowSeek reads depth with. LICENSE-FlowSeek and NOTICE-FlowSeek.md beside them are upstream's own.

File Original Licence
optical_flow/flowseek_t_ct.safetensors flowseek_T_CT.pth from mattpoggi/flowseek, Matteo Poggi and Fabio Tosi Apache-2.0
optical_flow/flowseek_t_tskh.safetensors flowseek_T_TartanCT_TSKH.pth, same Apache-2.0

models/kandinsky6/

Kandinsky 6 for ComfyUI, in ComfyUI's own folder layout: each subfolder goes into the matching folder under ComfyUI/models. The transformers are ComfyUI int8 ConvRot and W6A8 quantisations; everything else is byte for byte the upstream file. models/kandinsky6/NOTICE.md names each source commit and every SHA-256. The licence files sit beside the files they cover.

File Original Licence
models/kandinsky6/diffusion_models/kandinsky6_lite_distill_5s_int8_convrot.safetensors kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (int8_convrot) MIT
models/kandinsky6/diffusion_models/kandinsky6_lite_distill_5s_w6a8.safetensors kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (w6a8) MIT
models/kandinsky6/diffusion_models/kandinsky6_lite_5s_int8_convrot.safetensors kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (int8_convrot) MIT
models/kandinsky6/diffusion_models/kandinsky6_lite_5s_w6a8.safetensors kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (w6a8) MIT
models/kandinsky6/diffusion_models/kandinsky6_pro_distill_5s_int8_convrot.safetensors kandinskylab/Kandinsky-6.0-Pro-distill-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (int8_convrot) MIT
models/kandinsky6/diffusion_models/kandinsky6_pro_distill_5s_w6a8.safetensors kandinskylab/Kandinsky-6.0-Pro-distill-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (w6a8) MIT
models/kandinsky6/diffusion_models/kandinsky6_pro_5s_int8_convrot.safetensors kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (int8_convrot) MIT
models/kandinsky6/diffusion_models/kandinsky6_pro_5s_w6a8.safetensors kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers, Kandinsky Lab, quantised for ComfyUI (w6a8) MIT
models/kandinsky6/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors Qwen/Qwen2.5-VL-7B-Instruct, as repackaged by Comfy-Org Apache-2.0
models/kandinsky6/text_encoders/clip_l.safetensors openai/clip-vit-large-patch14, as repackaged by Comfy-Org MIT
models/kandinsky6/vae/kandinsky6_audio_vae.safetensors MMAudio's 44.1 kHz autoencoder and BigVGAN v2, as packed by Kandinsky Lab CC BY-NC 4.0, non-commercial (MMAudio); MIT (BigVGAN)

The video VAE Kandinsky 6 decodes with, hunyuan_video_vae_bf16.safetensors, is at Comfy-Org/HunyuanVideo_repackaged under the Tencent Hunyuan Community License.

models/bluepixel/

BluePixel-Turbo for ComfyUI, in ComfyUI's own folder layout: each subfolder goes into the matching folder under ComfyUI/models. The transformers are vivo's release converted to the FLUX.2 layout ComfyUI reads, at bfloat16 and as ComfyUI int8 ConvRot, W6A8 and W4A8 quantisations; the autoencoder is byte for byte the upstream file. models/bluepixel/NOTICE.md carries the FLUX Non-Commercial License attribution notice, names each source commit and every SHA-256. The licence files sit beside the files they cover.

File Original Licence
models/bluepixel/diffusion_models/bluepixel_turbo_bf16.safetensors openvivo/BluePixel-Turbo, vivo, a modified FLUX.2 [klein] 9B, converted for ComfyUI FLUX Non-Commercial License, non-commercial
models/bluepixel/diffusion_models/bluepixel_turbo_int8_convrot.safetensors the same, quantised for ComfyUI (int8_convrot) FLUX Non-Commercial License, non-commercial
models/bluepixel/diffusion_models/bluepixel_turbo_w6a8.safetensors the same, quantised for ComfyUI (w6a8) FLUX Non-Commercial License, non-commercial
models/bluepixel/diffusion_models/bluepixel_turbo_w4a8.safetensors the same, quantised for ComfyUI (w4a8) FLUX Non-Commercial License, non-commercial
models/bluepixel/vae/dc-ae-f64c128-mix-1.0.safetensors mit-han-lab/dc-ae-f64c128-mix-1.0-diffusers, MIT Han Lab Apache-2.0

The text encoder, qwen_3_8b.safetensors, is at Comfy-Org/Flux-2-Klein-9B under split_files/text_encoders/, with the same weights as vivo's.

Citations 

Zhang et al., Practical Blind Denoising via Swin-Conv-UNet and Data Synthesis, 2022.
Chen et al., Simple Baselines for Image Restoration, ECCV 2022.
Yan et al., HVI: A New Color Space for Low-light Image Enhancement, CVPR 2025.
Cai et al., Retinexformer: One-stage Retinex-based Transformer for Low-light
Image Enhancement, ICCV 2023.
Feijoo et al., DarkIR: Robust Low-Light Image Restoration, 2024.
Zheng et al., Bilateral Reference for High-Resolution Dichotomous Image
Segmentation, 2024.
Poggi and Tosi, FlowSeek: Optical Flow Made Easier with Depth Foundation Models
and Motion Bases, ICCV 2025.
Yang et al., Depth Anything V2, NeurIPS 2024.
Kandinsky Lab, Kandinsky 6, 2026.
Cheng et al., MMAudio: Taming Multimodal Joint Training for High-Quality
Video-to-Audio Synthesis, CVPR 2025.
Lee et al., BigVGAN: A Universal Neural Vocoder with Large-Scale Training, ICLR 2023.
Bai et al., Qwen2.5-VL Technical Report, 2025.
Radford et al., Learning Transferable Visual Models From Natural Language
Supervision, ICML 2021.
vivo, BluePixel: Toward Native 4K Agentic Image Editing, 2026.
Chen et al., Deep Compression Autoencoder for Efficient High-Resolution Diffusion
Models, ICLR 2025.
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support