Award listed
Original award wording
Overall 3
Project profile
Custom Universe
A photo-to-3D scene editing project intended to make it easier for robotics labs to construct synthetic training environments.
An official interview describes scanning objects, placing them in a scene, and editing by dragging. The current repository describes a Three.js scene whose viewport is captured after edits and redrawn by a FLUX image model; a TRELLIS service converts photos to GLB models.
The repository returns placeholder geometry when TRELLIS is not configured. Its default local models require substantial CUDA memory and access to restricted models, so this is not established as a low-setup browser tool. Precise distance placement was a future plan in the interview. No models were generated, and neither speed nor robotics-training suitability was validated.
Creator or team (original public names)
Jake Stevens; Mauricio Pereira (official result credits)
Links & sources
Links may have changed since we checked them. Descriptions from creators have not been independently tested.
Details in the original language
Original title
Custom Universe
Original creator credit
Jake Stevens、Mauricio Pereira（官方结果署名）
作品是什么
Custom Universe 是照片到 3D 场景的编辑项目。
给谁用
需要合成训练环境的机器人实验室。
问题或体验
降低搭建特定场景训练材料的门槛。
如何使用
官方访谈描述扫描物体、放入场景并拖动编辑；当前仓库另说明每次编辑捕获视口，再由后端图像模型重绘，未运行。
得到什么
可编辑的场景与重绘图像；未核实是否能直接交付合格机器人训练数据。
关键机制
当前仓库列出 Three.js 场景、FLUX 图像编辑和 TRELLIS 照片转 GLB 服务；TRELLIS 未配置时只返回占位模型。
特点
用 3D 摆放约束图像重绘；依据当前 README 归纳。
Events & award announcements
Read original announcement
Result listed — see original announcement
Original award wording
Finalists
Sources last checked: 2026-10-10