According to Dynamic Observation Beating monitoring, ByteDance's Seed team has released Seed3D 2.0, a 3D model generator that can turn a single image into a textured 3D asset. This upgrade focuses on geometric precision and material realism, and the API has been integrated into the Volcano Ark platform.
The geometric generation adopts a Coarse-to-Fine two-stage strategy: first establishing a coarse topology using a large-parameter DiT, then recovering sharp edges and fine surfaces. On the material side, the MoE architecture is used to enhance high-resolution details, introducing VLM priors to improve material decomposition stability under unknown lighting conditions. It outputs complete PBR maps that can be directly plugged into a standard rendering pipeline.
Sixty evaluators with 3D modeling experience were blinded tested with about 200 test cases. Seed3D 2.0 was compared pairwise withHunyuan3D-2.5/3.1, Tripo 3.0, Rodin Gen2, HiTem v2.0, and the previous generation Seed3D 1.0. The geometric generation preference rates ranged from 65.1% to 98.3%, while the textured 3D asset preference rates all exceeded 69%.
In terms of downstream tasks, Seed3D 2.0 can segment 3D assets into separate parts by function, add joint information, and output URDF-compatible formats for simulation engines like Isaac Sim, suitable for dynamic interaction scenarios such as robot grasping. At the scene level, it supports text, multi-view images, or video inputs, combining multiple assets to create a complete scene.
