DreamFields 3D

An open-source toolkit that "dreams" colourful 3D objects from text and image prompts, exporting meshes, turntable videos and NeRF instances.

Launch work ↗

DreamFields 3D is an open-source toolkit that runs on Colab. Given a text or image prompt, it optimises a neural radiance field (NeRF) under CLIP’s judgement until its renderings “look like” the prompt, then exports the result as an obj / ply mesh with vertex colour, a 360° video and multi-view images.

It was an experiment from 2022, when text-to-3D was still in its infancy. Adapted from Google Research’s dreamfields and dreamfields-torch, it adds mesh export, random-view training, image augmentation and a choice of CLIP models, and has collected more than four hundred stars on GitHub.

Example prompt: “a cyborg organic biological pavilion could breathe with building skin containing algae”.

Code and Colab notebook: github.com/shengyu-meng/dreamfields-3D


Series: Boundless Intelligence

Search