Generative AI Research Intern - Image-to-Video Diffusion
Summary
Research internship at Apple (Beijing) on image-to-video diffusion: designing and evaluating generative models for motion modeling and temporal consistency, building experiment pipelines, and running ablations and benchmarks. Core stack is Python and PyTorch, with video-quality metrics like LPIPS and FVD.
We are seeking a talented Generative AI Research Intern to work on image-to-video (I2V) diffusion models. This role focuses on building next-generation generative systems that transform still images into high-quality, temporally consistent videos. You will work closely with researchers and engineers to explore cutting-edge diffusion and transformer-based architectures for video generation, motion modeling, and temporal representation learning.
You will:
- Lead experimental studies on image-to-video diffusion techniques and temporal generative modeling.
- Design, implement, and evaluate model variants for motion modeling and temporal consistency.
- Build experimental pipelines and run ablation studies and benchmarking experiments.
- Analyze results and summarize findings in internal research reports or presentations.
- Study and reproduce recent work in video diffusion and I2V generation.
- Participate in research discussions and proposal of new ideas.
Minimum Qualifications
- Currently pursuing a Master’s or Ph.D. degree in Computer Science, Electrical Engineering, Applied Mathematics, or a related field.
- Strong background in computer vision and deep learning, with research or project experience in diffusion models, flow-matching, or transformer-based generative models applied to video generation or spatiotemporal tasks.
- Experience or solid understanding of image-to-video models, video generation, temporal modeling, or video representation.
- Proficiency in Python and PyTorch, with the ability to implement research code independently.
- Familiarity with model training workflows, ablation studies, and experimental result analysis.
- Familiarity with generative video evaluation metrics such as LPIPS, FVD, and perceptual assessment methods.
- Strong motivation for independent research and the ability to drive projects with minimal supervision.
- Ability to read, reproduce, and critically analyze recent research papers.
Preferred Qualifications
- Understanding of video fundamentals such as motion estimation, optical flow, temporal consistency, or alignment techniques.
- Prior experience with large-scale training, multimodal models, or video datasets.
- Publications in top-tier conferences or journals (e.g., CVPR, ICCV, ECCV, NeurIPS, ICLR, etc.).
- Strong curiosity and motivation for independent research in generative video modeling.
