Div]:bg-bg-000/50 [&_pre>div]:border-0.5 [&_pre>div]:border-border-400 [&_.ignore-pre-bg>div]:bg-transparent [&_.standard-markdown_:is(p,blockquote,h1,h2,h3,h4,h5,h6)]:pl-2 [&_.standard-markdown_:is(p,blockquote,ul,ol,h1,h2,h3,h4,h5,h6)]:pr-8 [&_.progressive-markdown_:is(p,blockquote,h1,h2,h3,h4,h5,h6)]:pl-2 [&_.progressive-markdown_:is(p,blockquote,ul,ol,h1,h2,h3,h4,h5,h6)]:pr-8"> _*]:min-w-0 standard-markdown"> What if we could give artists the same precise camera control in AI-generated video that Pixar has in rendered animation—without sacrificing the creative spontaneity of diffusion models? We're the ~50-person team behind Stable Diffusion, Stable Video Diffusion, and FLUX.1—models with 400M downloads. But here's the frontier we're exploring: generative models are incredibly powerful at creating visual content, yet they lack the precise spatial control that makes the difference between a happy accident and a deliberate creative choice. We need someone who can bridge the gap between 3D geometry and diffusion models. What You'll Pioneer You'll work on one of the most challenging problems in generative AI: teaching models that learned to create from pixels alone to understand and respect the mathematics of 3D space. This isn't about bolting camera controls onto an existing system—it's about fundamentally rethinking how diffusion models can internalize geometric constraints. You'll be the person who: Trains large-scale diffusion transformer models for camera-controllable image and video generation—pushing the boundaries of what's possible when generative models meet geometric precision Develops conditioning mechanisms that allow diffusion models to understand 3D camera parameters (poses, trajectories, intrinsics) as naturally as they currently understand text prompts Rigorously ablates design choices for 3D control, running experiments that tell us not just what works, but why—and communicating those insights to shape our research direction Reasons about the speed-quality tradeoffs of 3D-aware architectures in production settings where both matter Questions We're Wrestling With How do you condition a diffusion model on camera parameters without breaking what makes it powerful in the first place? What's the right balance between geometric precision and generative freedom—and does that balance change depending on the use case? Can we maintain multi-view consistency in generated videos without sacrificing temporal coherence? Where do 3D priors help diffusion models, and where do they get in the way? How do we make 3D-aware architectures fast enough for real-world use while maintaining the quality that makes them worthwhile? These aren't theoretical questions—we're actively building systems where the answers matter. Who Thrives Here You live at the intersection of classical 3D computer vision and modern generative AI. You understand projective geometry deeply enough to debug why a conditioning mechanism isn't respecting camera intrinsics, and you understand diffusion models well enough to train them at scale without them collapsing. You likely have: Hands-on experience training large-scale diffusion models for image and video data—the kind where you've debugged training instabilities at 3am and lived to tell the tale Strong foundations in 3D projective geometry, camera models, and coordinate systems (this isn't nice-to-have; it's essential) Experience with 3D-aware generative models or neural rendering techniques—NeRFs, 3D Gaussian Splatting, or related approaches A track record of integrating geometric priors and 3D conditioning into neural networks in ways that actually work Deep proficiency in PyTorch, transformer architectures, and the full ecosystem of modern deep learning Solid understanding of distributed training techniques—FSDP, low precision training, model parallelism—because our models don't fit on one GPU We'd be especially excited if you: Have wrestled with multi-view consistency in generative models and understand why it's harder than it looks Bring experience with camera calibration, structure-from-motion, or SLAM from classical computer vision Have published research at the intersection of 3D vision and generative models Understand the tradeoffs between different 3D representations and when to use which What We're Building Toward We're not just adding features—we're exploring fundamental questions about how generative models can understand space. Every experiment teaches us something about the relationship between geometry and generation. Every ablation study reveals assumptions we didn't know we were making. If that sounds more compelling than implementing existing techniques, we should talk. We're based in Europe and value depth over noise, collaboration over hero culture, and honest technical conversations over hype. Our models have been downloaded hundreds of millions of times, but we're still a ~50-person team learning what's possible at the edge of generative AI.