ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models

https://ubc-vision.github.io/vivid123/ View-conditioned Diffusion model (Zero-1-to-3 XL) ViVid Guidance Input image Target camera trajectory Target camera trajectory Figure 1 . Teaser -we present a strikingly simple training-free method to make already available novel-view synthesis diffusion models more consistent both in terms of the desired viewing angle and the content-combining it with video diffusion. As shown in the example, the results of our method are more consistent with the input images and correspond more to the target views.