SimWorld Benchmark: 35.8% of 3D Scene Edits Break Something Else
ZhitingHu · x · 2026-10-01
The SimWorld team releases SimWorld-Code4Scene: 190 Unreal Engine cases testing spatial reasoning via text-to-scene construction and precise control via image-guided editing. Unlike screenshot-based evaluation, they reopen the actual generated 3D scenes and assess instruction following, physical validity, and scene integrity. Astra leads overall, but spatial composition is the weakest construction category across all 14 tested configurations, and 35.8% of edits that fully recover their targets still introduce unintended changes elsewhere. Paper and code are available.
Related event: Code4Scene Benchmark Tests Agents Building 3D Scenes in Unreal Engine(2 posts)→
More from Multimodal
- Generating Video with Just a Click and Drag: New Interactive Demo — adamho · 2026-10-01
- Free OpenAI Dot Avatar Maker Runs in Browser: Pick Shape, Color, Get 3D Character in a Minute — yihui_indie · 2026-10-01
- Seedance 2.5 Prompt Recipe: A 20-Second First-Person High-Speed Descent Video — LudovicCreator · 2026-10-01
- Change one word: Midjourney rebuilds a salt cathedral's ruins with waves — tisch_eins · 2026-10-01
- ElevenReader rolls out Eleven v4: more expressive voices and language support jumps from 32 to 90+ — lukeharries · 2026-10-01
- Creatify's Model Playground locks first/last frames, mixes product, person, scene, audio references — testingcatalog · 2026-10-01