Oriel — AI & technology worth noticing, every day.

IFM/PAN2

AI model · https://huggingface.co/IFM/PAN2

First seen by Oriel on .

AI summary

PAN 2 is a general-purpose world model that predicts action outcomes as videos, combining structured language reasoning with a 29B-parameter latent diffusion backbone.

  • Given an optional image, observation text, and textual actions, PAN 2 generates 704x1248 video at 24 fps for up to 20 seconds.
  • Its language model writes a structured, time-grounded world-state description, while the diffusion model supplies perceptual detail conditioned on that description and the starting image.
  • The released diffusion checkpoint supports text-to-video and image-to-video, uses BF16, and can run through the repository’s generate.py, Diffusers, or vLLM.
  • The recommended pan2-guardrail reduced harmful T2V outputs from 33.9% to 1.5% internally, but blocked 11.5% of benign T2V requests.
  • Long plans often fail, prompt wording strongly affects results, appearance can drift across clips, and simulations may violate intermediate physical constraints.

#world models #video generation #latent diffusion #simulative reasoning

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 208, latest observation ; +49 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 374, latest observation ; +39 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsIFM/PAN2https://huggingface.co/IFM/PAN2 Read discussion