About the job
About Us
Lightricks is an innovative AI-driven company, dedicated to developing state-of-the-art content creation technologies for businesses, enterprises, and studios. Our mission is to seamlessly connect creativity with technology. At the heart of our efforts is LTX-2, an open-source generative video model designed to produce expressive, high-fidelity videos at unparalleled speed. LTX-2 powers our proprietary products and a rapidly expanding network of partner integrations via API.
Globally recognized for our pioneering work in consumer creativity, our flagship product, Facetune, stands as one of the most renowned creative brands, empowering hundreds of millions of users globally with AI-enhanced visual expressions. We blend deep research, user-centric design, and comprehensive execution from initial concept to final product, striving to redefine the future of creative expression.
The Position
In response to the success of LTX-2, our popular open-source text-to-audio+video model, we are intensifying our research efforts to innovate advanced audio+video generation models. We are seeking talented Research Scientists to join our LTX-Applications team.
As a Research Scientist within the LTX Model Quality team, you will be instrumental in enhancing the quality, controllability, and alignment of our video generation models. This position emphasizes the critical post-training phase, where you will develop and apply techniques such as preference optimization, reward modeling, and human feedback integration to fine-tune model outputs. You will establish robust evaluation frameworks, define quality metrics, and implement systematic methodologies to detect and resolve model shortcomings. Your contributions will directly influence the quality of videos we produce.
Your Responsibilities
- Develop and execute post-training pipelines, including Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimization (DPO), and other optimization techniques tailored for video generation models.
- Fine-tune and manage Very Large Language Models (VLLMs) for comprehensive video and audio understanding.
- Design and refine evaluation metrics and frameworks to assess quality.
- Conduct thorough analyses of failure modes and devise targeted strategies to close quality gaps.
- Curate and build high-quality preference datasets and evaluation benchmarks capturing the intricate dimensions of video generation quality.
- Collaborate closely with fellow researchers to maintain effective feedback loops between human evaluations and model enhancements.
Your Qualifications
- A Master's degree or equivalent practical experience in computer vision or generative AI.
- Proven experience with post-training techniques applicable to generative or multimodal models.
- Solid grasp of evaluation methodologies, quality metrics, and benchmark development for generative AI.
- Strong analytical skills and a problem-solving mindset.
- Exceptional collaboration and communication abilities.
