AI
Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 09:24:35 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 03:18:25 UTC.
Briefings
Physis-Lang is a self-evolving framework that treats physical language as a shared representation for video generation, replacing visual signals with optimized physical descriptions. By constructing PhysCapBench, the authors decompose physical processes into atomic assertions and evaluate captions using recall and precision. An agentic loop iteratively refines instructions to correct assertion-level errors, while language-guided retrieval identifies missing physical processes. Experiments on four benchmarks with Wan and Cosmos backbones show consistent improvements in physical plausibility, with enhanced models surpassing the leading proprietary Veo 3.1 model.
Key points
- Physis-Lang treats physical language as a shared and optimizable representation across data curation, model training, and video generation to replace visual signals.
- PhysCapBench decomposes physical processes into atomic assertions and evaluates captions using recall and precision to identify missing physical processes.
Context and limitations
Results are limited to four widely used physical video benchmarks; specific performance metrics for other benchmarks are not reported.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 09:24:35 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeigg5rbqu7ysg6ar4neiwapseq4ancfxnglqqe4at6dsvyju7fedte