Research Scientist – Controlled 3d Generation

Stability AI · AI Frontier · Remote · Research

Research Scientist focused on 3D generation using flow matching and diffusion models, aiming to create consistent, editable, and physically grounded 3D assets and scenes. Responsibilities include research, designing training pipelines, developing conditioning techniques, analyzing model behavior, and publishing results.

What you'd actually do

Conduct cutting-edge research on flow-matching, diffusion, and score-based methods for 3D generation and reconstruction.
Design and implement scalable training pipelines for controllable 3D generation (meshes, Gaussians, NeRFs, voxels, implicit fields).
Develop techniques for conditioning and control (text, sketch, pose, camera, physics) and multi-view consistency.
Analyse model behaviour through ablations, visualisations, and quantitative metrics.
Collaborate with cross-disciplinary research, graphics, and infrastructure teams to translate research into production-ready systems.

Skills

Required

PhD (or equivalent experience) in Machine Learning, Computer Vision, or Computer Graphics
PyTorch, JAX, or CUDA-level optimisation
Understanding of 3D representations (meshes, Gaussians, signed-distance fields, volumetric grids, implicit networks)
Solid grasp of geometry processing, multi-view consistency, and differentiable rendering

Nice to have

Experience generating coherent 3D scenes with multiple interacting objects, lighting, and spatial layout
Familiarity with scene-level control (object placement, camera path, simulation, or text-to-scene composition)
Knowledge of video-to-3D, image-to-scene, or 4D temporal generation
Background in physically-based rendering, simulation, or world-model architectures
Track record of impactful publications or open-source releases

What the JD emphasized

Published work on diffusion, flow-matching, or score-based generative models (2D or 3D)
PyTorch, JAX, or CUDA-level optimisation
3D representations (meshes, Gaussians, signed-distance fields, volumetric grids, implicit networks)
geometry processing, multi-view consistency, and differentiable rendering

Other signals

3D generation
flow matching
diffusion models
controllable 3D content creation

Read full job description

Research Scientist – Controlled 3D Generation

Location: Remote

About the Role

We’re seeking a Research Scientist passionate about 3D generation, flow matching, and diffusion models. You’ll help advance the frontier of controllable 3D content creation—building models that generate consistent, editable, and physically grounded 3D assets and scenes.

What You’ll Do

Conduct cutting-edge research on flow-matching, diffusion, and score-based methods for 3D generation and reconstruction.
Design and implement scalable training pipelines for controllable 3D generation (meshes, Gaussians, NeRFs, voxels, implicit fields).
Develop techniques for conditioning and control (text, sketch, pose, camera, physics) and multi-view consistency.
Analyse model behaviour through ablations, visualisations, and quantitative metrics.
Collaborate with cross-disciplinary research, graphics, and infrastructure teams to translate research into production-ready systems.
Publish results at top-tier venues and work with interns.

What You Bring

PhD (or equivalent experience) in Machine Learning, Computer Vision, or Computer Graphics.
Published work on diffusion, flow-matching, or score-based generative models (2D or 3D).
Strong engineering and problem-solving abilities: experience with PyTorch, JAX, or CUDA-level optimisation.
Understanding of 3D representations (meshes, Gaussians, signed-distance fields, volumetric grids, implicit networks).
Solid grasp of geometry processing, multi-view consistency, and differentiable rendering.
Ability to scale experiments efficiently and communicate complex results clearly.

Bonus / Preferred

Experience generating coherent 3D scenes with multiple interacting objects, lighting, and spatial layout.
Familiarity with scene-level control (object placement, camera path, simulation, or text-to-scene composition).
Knowledge of video-to-3D, image-to-scene, or 4D temporal generation.
Background in physically-based rendering, simulation, or world-model architectures.
Track record of impactful publications or open-source releases.

Equal Employment Opportunity:

We are an equal opportunity employer and do not discriminate on the basis of race, religion, national origin, gender, sexual orientation, age, veteran status, disability or other legally protected statuses.