Scenethesis: A Language and Vision Agentic Framework for 3D Scene Generation
Lu Ling, Chen-Hsuan Lin, Tsung-Yi Lin, Yifan Ding, Yu Zeng, Yichen Sheng, Yunhao Ge, Ming-Yu Liu, Aniket Bera, Zhaoshuo Li
2025-05-07
Summary
This paper talks about Scenethesis, a system that uses AI language models together with visual guidance to create detailed and realistic 3D scenes just from text descriptions.
What's the problem?
It is difficult for AI to generate 3D scenes that look real and make sense physically when only given a description in words, because understanding both language and visual details is very complex.
What's the solution?
The researchers combined language models that understand text with vision-based tools that refine the scene, so the AI can create diverse and believable 3D environments that match the descriptions accurately.
Why it matters?
This matters because it can make designing virtual worlds, games, and simulations easier and faster, allowing people to create realistic 3D scenes just by describing them in words.
Abstract
Scenethesis combines language models with vision-guided refinement to create diverse, realistic, and physically plausible 3D scenes from text.