Skip to content
RobothecaBeta
Log in

PhyWorld

  • AI model
  • Humanoids and robot learning

PhyWorld is a video-generation world model that continues a scene into physically plausible, temporally coherent frames, post-trained from Wan2.2-I2V-A14B.

Made by NU World Model & Embodied AI.

Source on GitHub: github.com/NU-World-Model-Embodied-AI/PhyWorld (opens in a new tab). Facts read at commit 38a61a0 (opens in a new tab).

Licenses

Code
Apache-2.0 (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit 38a61a0 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Fine-tunes Wan2.2-I2V-A14B with flow matching for video-to-video continuation, using a binary mask for preserved versus synthesized frames.
  • Applies Direct Preference Optimization over physics preference pairs drawn from the PhyGround human-annotation pool.
  • Uses a rank-16 LoRA adapter on the denoiser's attention and feed-forward projections during preference optimization.
  • Covers seven physical-event classes, including collisions, fluids, shadows and reflections, and ballistic throws.
  • Reports a PhyGround overall score of 3.09 and a VBench average of 0.769 in its comparison against baseline models.
  • The repository currently hosts the project page; training, inference, and evaluation code are promised later.

Written by Robotheca from the project's README (opens in a new tab) at commit 38a61a0.

Getting started

Difficulty
Advanced

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
21 May 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.