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 / models /kandinsky6 /NOTICE.md
WAS's picture
WAS
Kandinsky 6: replace raw transformers with int8_convrot and W6A8 quantisations
dcc864e verified
preview code
|
Raw History Blame Contribute Delete
8.92 kB

Kandinsky 6, mirrored

The files WAS Node Suite runs Kandinsky 6 with, laid out as ComfyUI's own models folder: diffusion_models/ for Load Diffusion Model, text_encoders/ for DualCLIPLoader with type kandinsky5, vae/ for Load VAE and Kandinsky 6 Audio VAE Loader. The text encoders and the audio VAE are byte for byte the upstream file named beside them; the transformers are ComfyUI quantisations of the upstream file, described below.

File Upstream Licence
diffusion_models/kandinsky6_lite_distill_5s_int8_convrot.safetensors int8 ConvRot of kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers at efbaeb9961770cc82fece5b9cb2197049d8c983c, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_lite_distill_5s_w6a8.safetensors W6A8 of kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers at efbaeb9961770cc82fece5b9cb2197049d8c983c, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_lite_5s_int8_convrot.safetensors int8 ConvRot of kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers at 5686616f04318cbec7b0a4978b4ff1697395de15, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_lite_5s_w6a8.safetensors W6A8 of kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers at 5686616f04318cbec7b0a4978b4ff1697395de15, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_pro_distill_5s_int8_convrot.safetensors int8 ConvRot of kandinskylab/Kandinsky-6.0-Pro-distill-5s-Diffusers at 1018d5828716e8bea3de99609f45c924de9c5564, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_pro_distill_5s_w6a8.safetensors W6A8 of kandinskylab/Kandinsky-6.0-Pro-distill-5s-Diffusers at 1018d5828716e8bea3de99609f45c924de9c5564, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_pro_5s_int8_convrot.safetensors int8 ConvRot of kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers at 295ca976810e318fb45c870ba4fc72c9a37fd548, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
diffusion_models/kandinsky6_pro_5s_w6a8.safetensors W6A8 of kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers at 295ca976810e318fb45c870ba4fc72c9a37fd548, transformer/diffusion_pytorch_model.safetensors MIT, Kandinsky Lab: diffusion_models/LICENSE-Kandinsky6
text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors Comfy-Org/Qwen-Image_ComfyUI at 1f12b17be14c89b026c51a91d67c32f84bb047bc, split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors, Comfy-Org's fp8 repackaging of Qwen/Qwen2.5-VL-7B-Instruct Apache-2.0, Alibaba Cloud: text_encoders/LICENSE-Qwen2.5-VL
text_encoders/clip_l.safetensors Comfy-Org/HunyuanVideo_repackaged at 533a43813f01517d9e624f9e824a85bb94094120, split_files/text_encoders/clip_l.safetensors, the text tower of openai/clip-vit-large-patch14 MIT, OpenAI: text_encoders/LICENSE-CLIP
vae/kandinsky6_audio_vae.safetensors kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers at efbaeb9961770cc82fece5b9cb2197049d8c983c, audio_vae/diffusion_pytorch_model.safetensors CC BY-NC 4.0, non-commercial, and MIT: see below

The transformers are ComfyUI quantisations

The Kandinsky 6 transformers are held here only as ComfyUI quantised checkpoints, made from the upstream file with comfy-kitchen's own quantisers and loaded by ComfyUI's Load Diffusion Model as they are.

Suffix Format
_int8_convrot int8_tensorwise with ConvRot rotation, group 256, one scale per output row
_w6a8 w6a8_int8, 6-bit weights in groups of 32 with ConvRot rotation, group 256, run with int8 activations

Quantised: the query, key, value and output projections and both feed-forward weights of every transformer block whose input width is a multiple of 256. Every other tensor is bfloat16. On Lite the audio stream is 896 wide, so its block weights stay bfloat16 too.

Kandinsky-6.0-Pro-5s-Diffusers is gated on Hugging Face behind an access form; its licence is MIT, the same as every other Kandinsky 6 transformer here.

Source transformers, SHA-256:

f84ea3266560bbd1f203a874b8cee43e080073820facaf3306cab8fae2798a97  kandinskylab/Kandinsky-6.0-Pro-distill-5s-Diffusers@1018d58287 transformer
b1b52668561d9138d93bb4da13c9999e615bf569be9f8cf65732d54652447f53  kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers@efbaeb9961 transformer
1550090ca56fed8c75dda0d30f8aa7b46775e8e1ebb086b70ad532e40eb878f9  kandinskylab/Kandinsky-6.0-Lite-5s-Diffusers@5686616f04 transformer
f0568ea71b9d754b9538808fc61662c22123fd852e9f7c667f6150ed5f725967  kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers@295ca97681 transformer

SHA-256 

b3fb6f0bd38077e769ef947555927a25d585269903328417bf8c6da8dda277c0  diffusion_models/kandinsky6_lite_distill_5s_int8_convrot.safetensors
e936e403761efd07463bc96a4d6608a8014399243c76a8b249d6e3b7be9398e9  diffusion_models/kandinsky6_lite_distill_5s_w6a8.safetensors
f6b8a302c99116ee1273a062b8978e10437a6b06f98c71f1c5ba49ecb485aa4c  diffusion_models/kandinsky6_lite_5s_int8_convrot.safetensors
375b16bd1f3981806d7f9513978f0c1e0a460d074d080fe1f267c189cbb895f4  diffusion_models/kandinsky6_lite_5s_w6a8.safetensors
8b96fc841b9ecd1d96218b70ecffd846ba8b620fa24d087fc99a547fbdc4edf9  diffusion_models/kandinsky6_pro_distill_5s_int8_convrot.safetensors
d26c82be04d5e2d97f4bcf184f0af09768f3af92ade3b7fff52ee181f0a0b968  diffusion_models/kandinsky6_pro_distill_5s_w6a8.safetensors
e21a0a4eeab4c25491d30b0af05358830cdd556ffe434c0866e41866c2ad694b  diffusion_models/kandinsky6_pro_5s_int8_convrot.safetensors
30ad0006e81bec43d00afa0c6c97b64dd94b47481287ec8b97d5e91a0ecab815  diffusion_models/kandinsky6_pro_5s_w6a8.safetensors
cb5636d852a0ea6a9075ab1bef496c0db7aef13c02350571e388aea959c5c0b4  text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors
660c6f5b1abae9dc498ac2d21e1347d2abdb0cf6c0c0c8576cd796491d9a6cdd  text_encoders/clip_l.safetensors
dfccdf9d0d8d3de270a23b097400f21adb91f4ec683bdc26369cb65b08a5af87  vae/kandinsky6_audio_vae.safetensors

The audio VAE is non-commercial

vae/kandinsky6_audio_vae.safetensors holds three parts, under the keys they load from:

Keys What it is Origin Licence
vae. The 44.1 kHz mel autoencoder ext_weights/v1-44.pth from hkchengrex/MMAudio, Sony Research Inc., with its weight normalisation folded into the stored weights CC BY-NC 4.0: vae/LICENSE-MMAudio-weights
vocoder. BigVGAN v2, 44.1 kHz, 128 bands, 512x bigvgan_generator.pt from nvidia/bigvgan_v2_44khz_128band_512x, NVIDIA, with its weight normalisation folded into the stored weights MIT: vae/LICENSE-BigVGAN
mel_converter. The mel filter bank and window MMAudio CC BY-NC 4.0

Every vae. tensor equals MMAudio's v1-44.pth with that folding applied, and every vocoder. tensor equals NVIDIA's generator with its own. Kandinsky Lab publishes the file inside its MIT repositories. MMAudio publishes its checkpoints under CC BY-NC 4.0, for non-commercial use only, and that licence is the one carried here. It is mirrored for non-commercial use, unchanged from Kandinsky Lab's file.

Not mirrored here

File Where it is Why
hunyuan_video_vae_bf16.safetensors, the video VAE Comfy-Org/HunyuanVideo_repackaged, split_files/vae/ Tencent Hunyuan Community License