PAN 2 is out with open weights: a 29B video model, clips up to 20 seconds at 720p/24fps, Apache 2.0
IFM/PAN2 · Hugging Face
Image-Text-to-Video model with 29B parameters. Download, explore files and run it on Hugging Face.
huggingface.co
Quick facts from the model page so nobody has to dig:
29B latent diffusion model
Up to 20 s per clip, 720p, 24 fps
Apache 2.0, so commercial use is on the table
There's a diffusers pipeline already, plus recipes for FastVideo and vLLM
What I want to know from anyone who has run it: real VRAM use at 720p with offloading, and whether 20 seconds actually stays coherent or falls apart after 8 like most "long" models.