Breaking

VAST-AI Releases AniGen, Turning a Single Image Into Animatable 3D Assets

July 4, 2026 at 05:23 EDT

Media Generation
Open Source
Research & Papers

VAST-AI-Research has released AniGen, a framework that generates animatable (rigged) 3D assets from a single image. With just one RGB image as input, it simultaneously produces a mesh, skeleton, and skinning weights that can be dropped directly into standard 3D pipelines.

SIGGRAPH 2026 · VAST-AI-Research

AniGen: One Image In, a Rigged 3D Asset Out

A research framework that generates animation-ready 3D assets — mesh, skeleton and skinning weights — directly from a single in-the-wild image, ready to drop into animation or physics simulation.

input image needed (in-the-wild)

S³

unified Shape · Skeleton · Skin fields

18GB

min. GPU VRAM (NVIDIA)

MIT

open license · research release

Two approaches to rigged 3D

CONVENTIONAL — sequential

Shape → Rig after

Rigging bolted on afterward — leaves gaps in rig consistency and animation quality.

ANIGEN — unified, end-to-end

Shape + Skeleton + Skin

All three modeled as mutually consistent S³ fields in one shared space.

From single image to animate-ready asset

Single image
animals · humanoids · machines

→

Two-stage
flow-matching
S³ fields

→

Rigged mesh (.glb)
+ skeleton visualization

Straight into embodied AI, physics simulation and character animation.

Claimed strength

Substantially surpasses sequential baselines in rig validity and animation quality, and generalizes to in-the-wild images.

Inherent limit

A single viewpoint means the back side and joints rely on estimation — there are limits to what one image can provide.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.