Open-Source LynnReal-Omni: One 32B Model for Video Gen, Editing and Restoration, 377ms on H100
AgeNo5351 · reddit · 2026-09-16
LynnReal-Omni is out: a 32B shared multimodal diffusion transformer built on the MiniMax H3 architecture, with open weights and ComfyUI nodes.
One model, many video tasks
- Text-to-video, image-to-video, human/hand pose-guided generation, structural control, omni-reference, style transfer, and video editing;
- Restores degraded video (reportedly also repairs accumulated-error artifacts);
- Streaming long-video generation;
- Accepts heterogeneous inputs (appearance refs, editable 3D renders, game recordings) so an agent can compose visual conditions in one model;
- Four-step fast generation.
Flash for real-time rendering
- A 27B three-step variant with lightweight VAE decoder;
- On a single H100, 22-frame 540p generation + decoding takes 843ms (Standard) or 377ms (Flash), laying groundwork for real-time streaming video.
Related event: LynnReal-Omni: Open-Source Unified Video Generation Framework(2 posts)→
More from Multimodal
- Audio8-TTS-Realistic-Air, a zero-shot voice-cloning TTS model, trends on Hugging Face — Edge0 · 2026-11-02
- Mistral explains why its model-built solver is a first step toward new drugs and batteries — MistralAI · 2026-11-02
- Mistral Large 4 writes a quantum chemistry SCF solver from scratch with one short prompt — MistralAI · 2026-11-02
- Gorgeous dithered pixel art collage serves up a 'great morning' — philipvollet · 2026-11-02
- Clown Prophecy moves to new repo, now uses first clip as reference to fix H3 quality drift — reeight · 2026-11-02
- A 25-year design veteran asks: who still mixes pro software with AI tools? — Opening-Knee-5913 · 2026-11-02