VIPER framework uses MLLM to improve physics in video generation

Researchers have introduced VIPER, a framework designed to improve the physical plausibility of generated videos. VIPER utilizes a Multimodal Large Language Model (MLLM) to extract physics-related cues from reference videos, which then guide a standard image-to-video generator. This approach allows for the transfer of physical behaviors, such as material response and motion trajectory, to new scenes without requiring complex text prompts. To support this, a new dataset called VIPER-19K has been created, featuring annotations for material properties, trajectories, and physical impacts. AI

IMPACT Enhances control over physical realism in AI-generated videos, potentially leading to more believable synthetic media.

RANK_REASON The cluster describes a new research paper detailing a novel framework and dataset for video generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Multimodal Large Language Model
• VIPER
• VIPER-19K

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →