Yoav HaCohen
I run the research team at LTX, where we build cutting-edge generative models for video, audio, and general visual intelligence. Our work spans the full stack—architecture design, training infrastructure, and open-source releases like LTX-Video and LTX-2. I founded LTX as a team inside Lightricks that grew into an independent company focused on foundational AI for generative media and physical AI.
My vision for pushing the boundaries of the frontier comes out of my experience in research and startups: a PhD and two companies I founded at the intersection of machine learning, computer vision, and generative AI.
Featured Projects
News
- Oct 2026 Paper APEX paper published—hierarchical latent prefixes and asynchronous denoising enable generation under extreme compression.
- Aug 2026 Release LTX-2.5 launched—text-to-video, image-to-video, and audio-to-video models supporting portrait and landscape formats.
- Jun 2026 Paper ScenA paper published—reference-driven multi-speaker audio scene generation from natural-language prompts.
- Mar 2026 Release LTX-2.3 released—22B parameter model with improved visual quality, native portrait support, and cleaner audio.
- Jan 2026 Paper LTX-2 paper and open-source weights published—the first production-ready open audio-visual foundation model, with native 4K output and synchronized sound.
- Nov 2025 Blog How Lightricks trains video diffusion models at scale with JAX on TPU—Google Cloud blog post with Yaki Bitterman.
- May 2025 Release LTX-Video 13B launched—multiscale rendering architecture generates video 30× faster than comparable models.
- Dec 2024 Paper LTX-Video paper published—realtime video generation via a transformer-based latent diffusion model.
Research Interests
Talks & Writing
Selected Publications
-
arXiv:2606.19325, 2026
-
Non-Rigid Dense Correspondence with Applications for Image EnhancementPh.D. Thesis, The Hebrew University of Jerusalem, 2014
-
ACM Transactions on Graphics (Proc. SIGGRAPH), 32(4), 2013
-
IEEE International Conference on Computer Vision (ICCV), 2013
-
ACM Transactions on Graphics (Proc. SIGGRAPH), 30(4), 2011
Full list on Google Scholar · DBLP
Experience
VP Research
Leading the GenAI Foundation Models group. Shipped LTX-Video, LTX-2, and LTX-2.3—open-source video and audio-visual generation models. Training 100B+ parameter models at scale from scratch, owning the entire stack—data, infra, architecture, pre-training, and post-training.
Chief Scientist & Founding Team
Deep-learning-based facial de-identification and creative reality technology.
Co-Founder & CEO
AI and computer-vision platform for analyzing, curating, and editing photo collections at scale—shipped as SDKs for app developers and through Picktorial, a top-rated professional photo editor for macOS.
Ph.D. in Computer Science
Dense correspondence algorithms for image enhancement. Advisor: Prof. Dani Lischinski. Published at SIGGRAPH and ICCV.