This list contains only the countries for which job offers have been published in the selected language (e.g., in the French version, only job offers written in French are displayed, and in the English version, only those in English).
At Luma, the Post-training team is responsible for unlocking creative control in the world’s largest and most powerful pre-trained multimodal models. The team works closely with the Fundamental Research team and the Product teams across Luma to train our image and video generative models improving their capabilities in the final step refining them to be better aligned with what our users expect.
Job Responsibility:
Optimize Luma's image and video generative models through targeted fine-tuning to improve visual quality, instruction adherence, and overall performance metrics
Implement reinforcement learning techniques including Direct Preference Optimization and Generalized Reward Preference Optimization to align model outputs with human preferences and quality standards
Partner closely with the Applied Research team to identify product requirements, understand diverse use cases across Luma's platforms, and execute targeted fine-tuning initiatives to address performance gaps and enhance user-facing capabilities
Conduct comprehensive side-by-side evaluations comparing model performance against leading market competitors, systematically analyzing the impact of post-training techniques on downstream performance metrics and identifying areas for improvement
Develop advanced post-training capabilities for Luma’s video models including Camera control, Object & character Reference, Image & Video Editing, Human Performance & Motion Transfer Approaches
Architect data processing pipelines for large-scale video and image datasets, implementing filtering, balancing, and captioning systems to ensure training data quality across diverse content categories
Research and deploy cutting-edge diffusion sampling methodologies and hyperparameter optimization strategies to achieve superior performance on established visual quality benchmarks
Research emerging post-training methodologies in generative AI, evaluate their applicability to Luma's product ecosystem, and integrate promising techniques into our Post-training recipe
Requirements:
Advanced degree (Master's or PhD) in Computer Science, Artificial Intelligence, Machine Learning, or related technical discipline with concentrated study in deep learning and computer vision methodologies
Demonstrated ability to do independent research in Academic or Industry settings
Substantial industry experience in large-scale deep learning model training, with demonstrated expertise in at least one of Large Language Models, Vision-Language Models, Diffusion Models, or comparable generative AI architectures
Comprehensive technical proficiency and practical experience with leading deep learning frameworks, including advanced competency in one of PyTorch, JAX, TensorFlow, or equivalent platforms for model development and optimization
Strong orientation toward applied AI implementations with emphasis on translating product requirements into technical solutions, coupled with exceptional visual discrimination and dedicated focus on enhancing visual fidelity and aesthetic quality of generated content
Proficiency in accelerated prototyping and demonstration development for emerging features, facilitating efficient iteration cycles and comprehensive stakeholder evaluation prior to production implementation
Established track record of effective cross-functional teamwork, including successful partnerships with teams spanning Product, Design, Evaluation, Applied, and creative specialists