Magic3D

Recent breakthroughs in text-to-image synthesis have been driven by diffusion models trained on billions of image-text pairs. Adapting this approach to 3D synthesis would require large-scale datasets of labeled 3D assets and efficient architectures for denoising 3D data, neither of which currently exist. In this work, we circumvent these limitations by using a pre-trained 2D text-to-image diffusion model to perform text-to-3D synthesis. We introduce a loss based on probability density distillation that enables the use of a 2D diffusion model as a prior for optimization of a parametric image generator. Using this loss in a DeepDream-like procedure, we optimize a randomly-initialized 3D model (a Neural Radiance Field, or NeRF) via gradient descent such that its 2D renderings from random angles achieve a low loss. The resulting 3D model of the given text can be viewed from any angle, relit by arbitrary illumination, or composited into any 3D environment.

We generate a 3D SDF and a texture field via two latent codes. We utilize DMTet to extract a 3D surface mesh from the SDF and query the texture field at surface points to get colors. We train with adversarial losses defined on 2D images. In particular, we use a rasterization-based differentiable renderer to obtain RGB images and silhouettes. We utilize two 2D discriminators, each on RGB image, and silhouette, respectively, to classify whether the inputs are real or fake. The whole model is end-to-end trainable. As several industries are moving towards modeling massive 3D virtual worlds, the need for content creation tools that can scale in terms of the quantity, quality, and diversity of 3D content is becoming evident. In our work, we aim to train performant 3D generative models that synthesize textured meshes which can be directly consumed by 3D rendering engines, thus immediately usable in downstream applications.