ByteDance — AI Creation & Innovation Team

A generative AI pipeline that transforms 1–2 uploaded images into fully navigable panoramic VR environments. Unlike traditional video generation, users don't just watch — they spatially explore and interact with AI-reconstructed immersive scenes through world model approaches.

An end-to-end pipeline that transforms a user's portrait into a fully animated 3D AR character — capable of perceiving and responding to the live camera environment in real time. The system spans 2D stylization, 3D mesh generation, and auto-rigging with 38 motion actions, deployed for mobile AR rendering. Characters contextually interact with the scene, enabling human-AI co-creation in short-form video.

Led optimization of Seedance 1.5, ByteDance's generative video foundation model — improving generation quality by 35–60% while reducing GPU inference cost by 25% through quality-latency tradeoff tuning. Productized and scaled core video capabilities including Video LoRA, start-end frame control, motion mimic, and audio-driven generation — reaching 1M+ daily creations and 21M+ views.

Deployed AI video features including Video LoRA and Photo-to-Singing to 100M+ users. Analyzed behavioral data on non-expert creator adoption, achieving 21M+ views and 610K+ submissions.