← sources

Open-H3-Regenerate-2K-IC

source: reddit (reddit:StableDiffusion)published 2026-11-09 06:22ingested 2026-11-10 02:40

retained text



[https://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K-IC](https://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K-IC)

[https://github.com/NVlabs/Sana/tree/sol-engine/models/Open-H3-Regenerate-2K/IC](https://github.com/NVlabs/Sana/tree/sol-engine/models/Open-H3-Regenerate-2K/IC)

New latent upscale model by NVIDIA-labs. Basically, it's v2. A few highlights compared to v1:

\- Face & Text are more consistent with the original video

\- Use H3 Ref2VA instead of fl2va.

\- Turbo LoRA, Regenerate LoRA, LMS LoRA are not merged. They also  use Hyperflow 8-step LoRA

\- New samplings: 3, 4, 8 steps.

\- Fast VAE-v2. The v1 fast VAE has grid artifacts in ComfyUI

document fetched 2026-11-10 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xrkcea/openh3regenerate2kic