Track4Gen: Teaching Video Diffusion Models to Track Points
Improves Video Generation

Hyeonho Jeong^1,2 Chun-Hao P. Huang¹ Jong Chul Ye² Niloy J. Mitra^1,3 Duygu Ceylan¹

¹Adobe Research ²KAIST ³UCL

Please click on other pages to see more results

Page 1 • Page 2 • Page 3 • Page 4

Please visit this page to see comprehensive (all four baselines) comparisons.

Video Generation Comparison

Pretrained Stable Video Diffusion vs Track4Gen

Input Image

Pretrained Stable Video Diffusion

Track4Gen

Video Generation Comparison

Finetuned Stable Video Diffusion vs Track4Gen

Input Image

Finetuned Stable Video Diffusion

Track4Gen

Video Generation Comparison

Track4Gen trained without Refiner vs Track4Gen

Input Image

Track4Gen without Refiner

Track4Gen