kandinskylab/KVAE-3D-2.0-t4s8

GitHub | Technical report | Habr article | KVAE-Video 2.0 t4s16

KVAE 2.0: Video tokenizer (t4s8)

KVAE-Video 2.0 t4s8 is the 4 x 8 x 8 variant of KVAE 2.0, a family of causal video tokenizers designed as latent representations for diffusion models. It compresses videos into continuous 16-channel latents and reconstructs them with high fidelity. Its fully convolutional architecture supports long videos through temporal block caching.

Model zoo

Model	Modality	Compression	Latent channels
KVAE-Image 2.0	Image	8 x 8	32
KVAE-Video 2.0 t4s8	Video	4 x 8 x 8	16
KVAE-Video 2.0 t4s16	Video	4 x 16 x 16	64

Inference

Run from the KVAE source repository root. The reference environment uses Python 3.11, PyTorch 2.8.0, and CUDA 12.8.

pip install -r requirements.txt
pip install --editable .

import torch

from data import VideoReader
from kvae import KVAEVideo

device = torch.device("cuda:0")
dtype = torch.bfloat16

model = (
    KVAEVideo.from_pretrained("kandinskylab/KVAE-3D-2.0-t4s8").eval().to(device=device, dtype=dtype)
)
reader = VideoReader(stream_pattern="*.png", input_norm="m11")
video = reader.read_video("path/to/video_frames")["frames"].unsqueeze(0)
video = video.to(device=device, dtype=dtype)

with torch.no_grad():
    latent = model.encode(video, seg_len=16).latent_dist.mode()
    reconstruction = model.decode(latent, seg_len=16).clip(-1, 1)

Temporal segments are processed through internal block caches. Do not interleave independent videos on the same model instance; use one KVAEVideo instance per concurrent stream.

Evaluation

Reconstruction was evaluated on MCL-JCV at 720p and BVI-DVC. The compact table below reports MCL-JCV; all compared models use 4 x 8 x 8 compression with 16 latent channels.

Model	PSNR↑	SSIM↑	LPIPS↓
HunyuanVideo 1.0	34.3	0.90	0.047
Wan 2.1	34.3	0.89	0.044
KVAE-Video 2.0 t4s8	36.0	0.92	0.047

Show reconstruction figures

Qualitative comparison

Columns from left to right: original video, KVAE-Video 2.0 t4s8, HunyuanVideo 1.0, and Wan 2.1.

Citation

@misc{kvae20,
  title         = {KVAE 2.0: video tokenizers for Image & Video generation models},
  author        = {Andrey Shutkin, Denis Parkhomenko, Kirill Chernyshev,
                  Ivan Kirillov, Denis Dimitrov,
                  Valeriya Kobenko, Kirill Malakhov},
  year          = {2026},
  eprint        = {2608.05798},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2608.05798}
}