Veo 3.1 (AI model): news timeline & CVEs

NVIDIA Researchers Introduce Physis-Lang: Self-Evolving Physical Language That Lifts Cosmos 3 Past Veo 3.1 on Physics Benchmarks

NVIDIA's Physis-Lang physics captions lift Cosmos 3 past Veo 3.1 on several physics video benchmarks.

NVIDIA, MIT, and University of Oxford researchers introduce Physis-Lang, a self-evolving natural-language representation of physics used for video-data curation, training captions, and inference prompts. On the September 29, 2026 Physics-IQ Verified snapshot, Cosmos3-Super with Physis-Lang scored 48.2 and Cosmos3-Nano scored 43.3; Cosmos3-Nano also beat Veo 3.1 on PhyGenBench (71.04 vs 65.63) and Physics-IQ Verified (43.41 vs 34.99) after LoRA fine-tuning. Distilling the captioning pipeline into Qwen3-VL-4B-Instruct models preserved most of a 7.05-point Wan2.1-14B gain at roughly $120 instead of about $24,120 in API cost. The comparison lists the work as paper-only as of September 30, 2026.

MarkTechPostSep 30, 07:32 UTC · 8d agoAI research

Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model

Physis-Lang improves physical video generation, and Cosmos3-Nano variants surpass proprietary Veo 3.1.

Physis-Lang represents physical processes in language covering entities, causes, interactions, principles, evolution, and effects, then optimizes that language across curation, training, and generation. PhysCapBench decomposes processes into atomic assertions and scores caption recall and precision, while an agentic loop fixes assertion errors and retrieves videos for missing processes. On four physical-video benchmarks, Wan and Cosmos backbones improve, and Cosmos3-Nano variants enhanced by Physis-Lang surpass proprietary Veo 3.1.

Hugging Face daily papersSep 29, 20:00 UTC · 9d agoAI research1

Google Research Introduces an AI Video Co-Director: 4 Agentic Frameworks for Coherent, Minutes-Long Video Generation

Google Research unveiled four agentic frameworks that generate coherent minutes-long video on Gemini and Veo.

Google Research introduced an AI video co-director of four agentic frameworks—Co-Director, CANVAS, A²RD, and VQQA—that orchestrate Gemini and Veo to turn short clips into coherent multi-shot videos lasting up to 10 minutes. The layer targets identity drift and cascading errors with bandit-based planning, persistent visual memory, training-free autoregressive diffusion, and closed-loop prompt refinement, and outputs inherit SynthID watermarking. Co-Director scored 81.4 on GenAD-Bench; CANVAS gained 21.6% background continuity; A²RD improved consistency by up to 30% on 1–10 minute videos. Co-Director and A²RD code is public; CANVAS is forthcoming and the stack is not a Google product.

MarkTechPostSep 28, 02:44 UTC · 10d agoAI research1

Veo 3.1

Timeline3 articles

Appears with

NVIDIA Researchers Introduce Physis-Lang: Self-Evolving Physical Language That Lifts Cosmos 3 Past Veo 3.1 on Physics Benchmarks

Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model

Google Research Introduces an AI Video Co-Director: 4 Agentic Frameworks for Coherent, Minutes-Long Video Generation