The project uses image diffusion to generate multiple images with multi-views of the same object. Diffusion models have been trained on a wide array of images given them highly accurate 3d understandings of the world. Using this novel technique, we can generate the vast number of images needed for gaussian splatting and thus generate a 3d model. Then, this can be viewed and downloaded as well as used as a new input modality for video editing, image editing, etc. with precise camera control provided by 3d model.
Research paper on gaussian splatting: https://arxiv.org/abs/2308.04079 PixVerse: https://app.pixverse.ai/
DEMO URL: https://drive.google.com/file/d/1Vh1JrPttGJib9gimeSROImGKpUiCtUiY/view?usp=drive_link
Built With
- pixverse
Log in or sign up for Devpost to join the conversation.