Hugging Face's logo

xin1u
/
OmniVR

Video-to-Video
PEFT
English
video-restoration
audio-video
audio-enhancement
omnivr-flash
lora
ltx-video
Model card Files
xet
 Community
OmniVR / ckpt /tinydecoder /README.md
Xin Lu
Update OmniVR model card and publish both inference paths
83aad91
preview code
|
Raw History Blame Contribute Delete
777 Bytes

Optional legacy TinyDecoder

taeltx2_3_wide.pth is the decoder-only TAEHV checkpoint from the earlier OmniVR release. Its implementation is bundled in inference_tiny_decoder.py. Attribution and its MIT license are retained in NOTICE.

The multistep entry point uses the full LTX video VAE by default. To select the legacy decoder explicitly:

hf download xin1u/OmniVR tinydecoder/taeltx2_3_wide.pth --local-dir ckpt
# Add this option to the multistep command:
# --tiny_decoder ckpt/tinydecoder/taeltx2_3_wide.pth

This checkpoint cannot replace the Flash Turbo VAE: Flash requires both an UltraTinyVAE encoder and decoder matching the bundled configs/turbo_vae.json.