UT Austin, Stanford et al. release unified survey of LLM-driven agentic computer graphics
UTEXAS · hf · 2026-11-03
Researchers from UT Austin, Stanford, Simon Fraser, UC Berkeley and Cambridge released "Advances in Agentic Computer Graphics", a survey proposing a unified taxonomy and literature system for LLM-driven agentic graphics. It decomposes agentic graphics systems into specification, harness, base model, graphics interface & tooling, and final output, covering 2D shapes (SketchAgent, IntroSVG), 3D static assets (3D-GENERALIST, BlenderGym), and 3D dynamic assets (Artcraft, 4DCodeBench), while analyzing trends and open challenges in human–agent collaboration and long-horizon planning.
More from Multimodal
- Map2Video uses maps and street view imagery to direct AI video of real locations — ryosuzk · 2026-11-04
- Open-source Reshoot 4D turns any video into a 4D Gaussian splat, runs on a 3090 — Scobleizer · 2026-11-04
- BobHead: open-source ComfyUI app turns selfie videos into big-head animations — jalbust · 2026-11-04
- 'Street birds' AI art series: dapper anthropomorphic birds painted in retro oil style — OVolosin82152 · 2026-11-04
- A 40-second AI-made ad for a free app, built with Seedance and Nano Banana — leonardsouza · 2026-11-04
- AI-illustrated fairy tale 'The Flower of the Night Watch' showcases cinematic AI art — LudovicCreator · 2026-11-04