PhD Intern – Generative AI Models

Job Description

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

The Tone:
This is a PhD internship at TikTok, with potential locations including Los Angeles. TikTok is the leading destination for short-form mobile video, focused on inspiring creativity and bringing joy. This internship offers students the opportunity to actively contribute to the company’s products, research, future plans, and emerging technologies, specifically in the critical area of privacy-focused multimodal generative AI. This role is essential for building trusted technology innovation that respects user privacy choices while advancing state-of-the-art generative models.

The TL;DR
• Role: Internship
• Location: Not specified
• Pay: $60 hourly
• Team: Privacy Innovation (PI) Lab – Multimodal Generative Model
• Mission: Develop next-generation, privacy-sensitive generative foundation models, focusing on diffusion-based and unified generation-understanding architectures for production environments.
• Tech Stack: PyTorch, GPU architecture, Diffusion, DiT, Flow Matching, Rectified Flow

What You’ll Actually Do
• Design and Optimize: Participate in the architecture design and deep optimization of next-generation text-to-image and text-to-video models.
• Develop and Implement: Lead or contribute to the design and implementation of Diffusion Transformer (DiT / MM-DiT) architecture improvements and unified multimodal model designs.
• Optimize Performance: Perform joint algorithmic and system-level optimization, targeting training stability, memory and compute efficiency, and generation quality.
• Address Challenges: Address challenges in long-sequence, high-resolution, and video generation, including efficient attention and temporal modeling strategies.
• Advance Research: Reproduce, analyze, and advance state-of-the-art generative models beyond simple replication.

The Must-Haves
• Background: Currently pursuing a PhD in Computer Science, Computer Engineering, or a related technical discipline.
• Experience: Hands-on experience training large-scale models, with practical experience in text-to-image or text-to-video models (non-toy systems) preferred.
• Skills: Deep understanding of Diffusion / Flow Matching / Rectified Flow models; strong familiarity with DiT / Transformer-based architectures in generative modeling; proficiency with PyTorch; ability to debug the full pipeline from mathematical formulation to generated outputs.
• Bonus: Familiarity with multimodal modeling (Text / Image / Video / Audio); research publications or open-source contributions.