ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models
This addresses the challenge of creating photorealistic 3D assets with backgrounds for applications in content creation, though it is incremental as it builds on existing diffusion models.
The paper tackles the problem of generating 3D-consistent images from text using pretrained text-to-image models, resulting in improved visual quality with reductions of 30% in FID and 37% in KID scores compared to existing methods.
3D asset generation is getting massive amounts of attention, inspired by the recent success of text-guided 2D content creation. Existing text-to-3D methods use pretrained text-to-image diffusion models in an optimization problem or fine-tune them on synthetic data, which often results in non-photorealistic 3D objects without backgrounds. In this paper, we present a method that leverages pretrained text-to-image models as a prior, and learn to generate multi-view images in a single denoising process from real-world data. Concretely, we propose to integrate 3D volume-rendering and cross-frame-attention layers into each block of the existing U-Net network of the text-to-image model. Moreover, we design an autoregressive generation that renders more 3D-consistent images at any viewpoint. We train our model on real-world datasets of objects and showcase its capabilities to generate instances with a variety of high-quality shapes and textures in authentic surroundings. Compared to the existing methods, the results generated by our method are consistent, and have favorable visual quality (-30% FID, -37% KID).