@aisearchio Hope it’s open source. It was amazing to see what the community built around MiniMax H3 — Fast H3, Sol Lightning, Hyperflow, etc. Would be cool to see the same here.
Little known thing: if you're trying to quickly evaluate media model quality, skin texture and sheen is a quick cheat to spot the heavily optimized ones.
Here's some first frames from video models from us, competitors, and off hugging face too. Ours has the best skin. 😤
Open-source video generation is now faster than playback without compromising quality.
Introducing Video Delta Net (VDN): hybrid attention for live text-to-video with near-lossless quality.
VDN accelerates Minimax-H3 by 75 - 90 x, generating 14 seconds of 768p video in 11 seconds on 8× NVIDIA B200 GPUs.
Checkpoints + training/inference code + Technical Blog ⬇️
(1/6)