PozzettiAndrea/ComfyUI-SAM3DBody

What it solves

This project provides a way to recover a full-body 3D human mesh from a single image. It integrates Meta's SAM 3D Body model into the ComfyUI ecosystem, allowing users to generate 3D human representations from 2D images.

How it works

It acts as a wrapper for the SAM 3D Body library, bringing its inference capabilities into ComfyUI as custom nodes. The project also includes enhancements such as support for multiple people in a single image and experimental depth-supported size correction to address issues where smaller people are incorrectly predicted as normal-sized people further away.

Who it’s for

This is for 3D artists, researchers, and ComfyUI users who want to automate the creation of 3D human meshes from photographs.

Highlights

• Single-Image Recovery: Generates 3D human meshes from a single 2D image.
• Multi-Person Support: Capable of processing multiple people within one image.
• ComfyUI Integration: Integrated as a set of custom nodes for node-based 3D workflows.
• Size Correction: Includes experimental depth-based corrections to improve the scale of predicted human figures.