Seed3D 1.0 is a 3D generation model from the ByteDance Seed team. It is a foundation model, a large neural network trained on broad data so that it can serve many tasks, and its task is to turn one input image into a 3D asset: a mesh (the geometric surface of an object) with textures and physically based rendering (PBR) materials, which describe how a surface reflects light.
The model generates the geometry and the textures, including multi-view images and PBR materials. A vision-language model (VLM), a model that reads both images and text, estimates the real-world size of each object, so the asset can be imported into physics simulators such as Isaac Sim, which builds collision meshes from the watertight geometry for robot grasping tests. For a whole scene, a VLM finds the objects in an image and their positions, Seed3D generates each object, and the scene is assembled from them, from offices to large urban scenes.
A technical report on the model appeared on arXiv in October 2025. The model is offered on the Ark console of volcengine.com; the downloaded pages show no price.