Research Scientist Intern (TikTok-Privacy Innovation Lab-Multimodal Generative Model) - 2026 Start (PhD)

Work on end-to-end training architecture for diffusion-based, unified text-to-image and text-to-video foundation models in privacy-sensitive, production settings. You’ll design and deeply optimize DiT and flow-matching/rectified-flow generative models, improve diffusion transformer architectures, and advance latent/tokenization/conditioning mechanisms. Focus on training stability, convergence, memory/compute efficiency, and generation quality, collaborating with systems and kernel engineers while reproducing and pushing state-of-the-art research.