Xin Chen

Chen received a BSc in electronic information science from Qingdao University of Technology in 2016. He studied computer science as a master’s student at ShanghaiTech University from 2016 to 2018 and completed his computer science doctorate at the University of Chinese Academy of Sciences in 2022. He worked as a research scientist at Tencent from February 2022 to June 2024, then joined ByteDance in July 2024.[2]

Controlling video generation

Chen co-authored the Seedance 2.0 model card. The model generates audio and video together and accepts text, images, audio and video as references. This lets a user guide a clip through several kinds of input rather than describing everything in words. His CV describes work on prompt processing, training and large-scale video captioning for Seedance, including models that turn a user’s instructions into more structured generation guidance.[2][3]

Generating 3D structure

MeshXL represents a 3D mesh, an object made from connected vertices and faces, as a sequence that can be modeled with methods drawn from language modeling. Its neural coordinate representation supports both generating a shape and completing a partial mesh. The project provides model weights, code and text-to-mesh examples, connecting research on geometric representations to tools people can run. Chen is a co-author.[1][4]

Adding textures without baked-in lighting

Paint3D takes an untextured mesh and a text or image prompt, then generates a texture map. Its coarse-to-fine process aims to remove lighting artifacts and cover regions that individual rendered views miss. Keeping illumination out of the texture makes the resulting object more useful when it is rendered under different lighting. Chen is an equal-contribution co-author of the project.[5]