Oriel — AI & technology worth noticing, every day.

Efficient-Large-Model/Open-H3-Regenerate-2K

AI model · https://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K

First seen by Oriel on .

AI summary

Open-H3-Regenerate-2K is an open replacement for the closed H3-Regenerate-2K API that refines 768P H3 drafts into 2560×1440 video using three MiniMax-H3 steps.

  • It processes 1344×768, 121-frame drafts with latent upscaling, three Euler refinement steps, CFG 1, and VAE decoding.
  • On one B200, the optimized pipeline takes 22.45 seconds for a five-second clip, versus 87.26 seconds with standard forward execution.
  • The fast H3 VAE decoder is a non-causal 24-layer decoder with 1.62B parameters that reduces decoding from 15 seconds to 1.6 seconds.
  • A merged LoRA combines three adapters across 312 linears, matching separate adapter deltas with a relative error of 2.2e-8.
  • The bundle uses multiple component licenses, including the MiniMax H3 Community License, rather than one blanket license.

#video generation #minimax-h3 #lora #vae

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 219, latest observation ; +107 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 483,649, latest observation ; +483,649 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsEfficient-Large-Model/Open-H3-Regenerate-2Khttps://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K Read discussion