Open-H3-Regenerate-2K-IC regenerates a MiniMax H3 draft at 1080p, 2K or 4K while keeping its faces, lip sync and lettering
Open-H3-Regenerate-2K-IC, released on 9 November 2026 by the NVlabs SoL-Engine team, is an in-context LoRA for MiniMax H3 that regenerates a low-resolution draft at 1080p, 2K or 4K in 3, 4 or 8 steps. It is an open alternative to MiniMax's H3-Regenerate-2K, which is only a hosted API, and it keeps the draft in context so faces, lip sync, text and layout stay as they were.
What it does
The draft stays in H3's Ref2VA transformer as a reference video and the LoRA generates the high-resolution version of the same clip; it also takes 384P, 512P or 576P drafts up to 768P. The default 3-step sampler enlarges the draft with a 3D latent upscaler and then runs the last steps of the HyperFlow 8-step grid with CFG off; you pass the original H3 prompt unchanged, or none.
Worth knowing
The weights are under the MiniMax H3 Community License, which excludes the European Union, the United Kingdom, South Korea and the United States (users there must request a licence from MiniMax), and it is not an official MiniMax product. The official script runs with torchrun on 1 to 8 GPUs, and the project page's speed figure (a 5-second 2K clip in 22.45 s) is for the earlier non-IC version on one NVIDIA B200; the page says that version could drift on faces, mouths and lettering, which the IC version fixes. A community ComfyUI conversion with a node pack and workflow (slmonker) appeared the same day.