From 787abc0be96c05e6585aea7348b52e1626c00672 Mon Sep 17 00:00:00 2001 From: Chun-Han Yao Date: Mon, 3 Nov 2025 21:59:51 +0000 Subject: [PATCH] initial SP4D commit --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 714e85c..1a20705 100755 --- a/README.md +++ b/README.md @@ -7,8 +7,8 @@ **Nov 4, 2025** - We are releasing **[Stable Part Diffusion 4D (SP4D)](https://huggingface.co/stabilityai/sp4d)**, a video-to-4D diffusion model for multi-view part video synthesis and animatable 3D asset generation. For research purposes: - - **SP4D** was trained to generate 48 frames (4 video frames x 12 camera views) at 576x576 resolution, given a 4-frame input video of the same size, ideally consisting of white-background images of a moving object. - - Based on our previous 4D model [SV4D 2.0](https://huggingface.co/stabilityai/sv4d2.0), **SP4D** can simultaneously generate multi-view RGB videos as well as the corresponding kinematic part segmentations that are consistency across time and camera views. + - **SP4D** was trained to generate 48 RGB frames and part segmentation maps (4 video frames x 12 camera views) at 576x576 resolution, given a 4-frame input video of the same size, ideally consisting of white-background images of a moving object. + - Based on our previous 4D model [SV4D 2.0](https://huggingface.co/stabilityai/sv4d2.0), **SP4D** can simultaneously generate multi-view RGB videos as well as the corresponding kinematic part segmentations that are consistent across time and camera views. - The generated part videos can then be used to create animation-ready 3D assets with part-aware rigging capabilities. - Please check our [project page](https://stablepartdiffusion4d.github.io/), [arxiv paper](https://arxiv.org/pdf/2509.10687) and [video summary](https://www.youtube.com/watch?v=FXEFeh8tf0k) for more details.