Dataset · AidanYuzhe
Coding agent benchmark for creating 4D scenes in Blender
AidanYuzhe/vibeprevisbench
Test suite for evaluating AI agents that turn scripts or photos into animated 4D scenes in Blender.
What's inside
The dataset includes fewer than a thousand English-language text and video examples for evaluating 3D agentic programming. The models' task is to generate executable Blender 4D scene code from a script, photos, or initial frame, including object placement, movement, framing, and editing cuts.
What you can do with it
- Evaluate coding agents in 3D/4D animation generation and previs
- Test multimodal LLMs on writing Blender scripts
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 2K
- Likes
- 3
How to load it
from datasets import load_dataset
ds = load_dataset("AidanYuzhe/vibeprevisbench")