Matan Ben Yosef

London, UK

Matan Ben Yosef

Generative-AI researcher and engineer at LTX (Lightricks). I enjoy training models and building cool stuff with them.

About

I'm a generative-AI researcher and engineer at LTX, Lightricks' video model team. I work on post-training, which mostly means taking a base video model and making it directable: control adapters, IC-LoRAs, and a lot of performance work on the training stack. I designed and built the LTX Trainer from scratch: the open-source trainer that ships with LTX-2, and what the community uses to fine-tune it. Earlier at Lightricks I built LoRA training infrastructure for SD and Flux, and the backend of a text-to-image product serving millions.

The research side of my work has followed one arc for a decade: generative models of people, from GANs to video diffusion. It started with my MSc at the Hebrew University of Jerusalem (2015–2018), on multi-modal GANs with Daphna Weinshall. I was at D-ID from 2018 to late 2021, ending as lead deep-learning and computer-vision researcher, where I built the face-animation models behind Live Portraits and AI Avatars and worked on adversarial face de-identification.

After D-ID I built FaceFX on my own: an iOS and Android app with around fifty face effects, each one its own GAN, served from GPU nodes on GCP, down to the ads and the subscriptions. Lightricks acquired it in mid-2022, and that is how I ended up here.

Before AI I spent about a decade as a software engineer, mostly on real-time and distributed systems. I've been writing code since I was ten, and I still care most about the same thing: simple, well-designed software.

Publications

  1. AVControl: Efficient Framework for Training Audio-Visual Controls

    Matan Ben-Yosef, Tavi Halperin, Naomi Ken Korem, Mohammad Salama, Harel Cain, Asaf Joseph, Anthony Chen, Urska Jelercic, Ofir Bibi arXiv preprint · 2026 arXiv Project

  2. JUST-DUB-IT: Video Dubbing via Joint Audio-Visual Diffusion

    Anthony Chen, Naomi Ken Korem, Gal Zeevi, Tavi Halperin, Matan Ben Yosef, Urska Jelercic, Ofir Bibi, Or Patashnik, Daniel Cohen-Or SIGGRAPH 2026 arXiv Project

  3. LumiVid: HDR Video Generation via Latent Alignment with Logarithmic Encoding

    Naomi Ken Korem, Mohamed Oumoumad, Harel Cain, Matan Ben Yosef, Urska Jelercic, Ofir Bibi, Yaron Inger, Or Patashnik, Daniel Cohen-Or arXiv preprint · 2026 arXiv Project

  4. LoRA Training for Text-to-Video Models: A Practical Guide to Fine-Tuning SOTA Video Generation

    Naomi Ken Korem, Matan Ben Yosef, Tavi Halperin, Ofir Bibi SIGGRAPH Asia 2025 ACM

  5. Gaussian Mixture Generative Adversarial Networks for Diverse Datasets, and the Unsupervised Clustering of Images

    Matan Ben-Yosef, Daphna Weinshall arXiv preprint · 2018 arXiv

  6. Multi-Modal Generative Adversarial Networks

    Matan Ben-Yosef MSc thesis · Hebrew University of Jerusalem · 2018 PDF

Open source

One production training system, and two command-line tools I built because I kept needing them.

Patents

Co-inventor on eight patents from my years at D-ID, in synthetic-media animation, anonymization, and face recognition.

Animation and reenactment

Anonymization and identity