Linum v2 - 2B parameter, Apache 2.0 licensed text-to-video models (360p, 720p)
linum.aiAny info on VRAM requirements and latency for the 720p model? With only 2B parameters it seems like it should be quite fast I guess.
The article actually notes it takes about 15 minutes for a 5-second clip on an H100. Video diffusion steps are much heavier than single-image inference, so the low parameter count doesn't necessarily translate to low latency.