David Dinkevich

prof_pic.jpg

I’m a Machine Learning researcher and MSc student at the Hebrew University of Jerusalem, where I work under the supervision of Dani Lischinski. My research focuses on generative models for images and video.

My current work centers on diffusion-based visual world models—systems that represent and simulate environments with applications in visual planning, animation, and AI-driven creativity. I am particularly interested in autoregressive video diffusion models, and in methods for generating long-duration, high-quality video efficiently.

I have experience modifying the internals of existing models—for example, my thesis explores adapting attention mechanisms in transformer-based image models to improve character consistency across frames.

I am motivated by a desire to open the black box and better understand how efficiently modern generative models process information.

selected publications

  1. Story2Board: A Training-Free Approach for Expressive Storyboard Generation
    David Dinkevich, Matan Levy, Omri Avrahami, and 2 more authors
    Under review, 2025
    arXiv: coming soon