Research Scientist / Engineer - Controllability, Personalization & Productization

You'll own the control interface of Luma's video foundation models: the fine-tuning, adapters, and personalization that turn a general model into something a top-tier creative partner can direct precisely. You sit between research, product, and partnerships, closing the gap between what the model can do and what production actually needs. You'll work fullstack across modeling, data, systems, and evaluation, with a bias toward the personalization and long-horizon memory that make a model feel like it understands a specific workflow. It fits a researcher who treats partners as collaborators and is drawn to specific, high-fidelity problems rather than leaderboard numbers. If public benchmarks are what motivate you most, this won't be the fit. What You'll Own Give Luma's models precise control and creative range for high-fidelity, partner-grade workflows using SFT, RL, distillation, and adapter-based methods. Embed domain expertise and long-horizon memory into the models with context management, PEFT, and preference learning. Ground the data engine in real-world workflows, turning usage into training and data insights that shape future models. Define and drive end-user quality: set success metrics, build user-aligned evaluations, and iterate the model/data/eval loop to strict fidelity targets. Partner with Product and Design to turn creative intent and feedback into clear specs and shippable model behaviors. First 90 Days One way the first 90 could unfold. Days 1–30 — Immerse & Diagnose: Learn the models and a priority partner's workflow, and pinpoint where control or personalization falls short today. Days 30–60 — Ship & Validate: Deliver a personalized or controllable variant that measurably improves a real partner workflow, with evaluation to back it. Days 60–90 — Scale & Systemize: Turn the approach into a repeatable adaptation loop that new verticals and partners can use. What You Bring Strong ML foundation with deep experience in visual generative models (diffusion, transformers, or related). Depth in at least one of: fine-tuning, personalization, domain adaptation, data curation, targeted distillation, interpretability, or human-feedback refinement. Hands-on experience with PyTorch and large model training. A product instinct: you treat users and partners as collaborators and solve their specific problems. Nice to Have Contributions to state-of-the-art image or video generation models. Experience working with creative partners (VFX, animation, film, design tools). A track record building workflows or tools that improve iteration speed and evaluation rigor. Familiarity with large-scale training infrastructure and distributed systems (Ray, Slurm, Kubernetes). About Luma: Luma's mission is to build unified general intelligence that can generate, understand, and operate in the physical world. We believe multimodality is critical for intelligence — the next step beyond language models comes from vision. Luma is an equal opportunity employer.