S
Generative AI Engineer
Spectronn
2 hours ago
Full-time
On-site
New Jersey, United States
Generative AI Engineer — Simulation & Generative Video
Must be a U.S. Citizen or Permanent Resident
About the role
Spectronn is hiring an engineer to lead synthetic video generation. The goal is to produce photorealistic, accurately labeled aerial video used to train and evaluate computer vision models. The ideal candidate can combine 3D model rendering with generative AI to achieve photorealism.
Spectronn is a tech startup building an agentic AI platform for multiple industries. You'll work directly with the technical co-founder/CTO and interact with customers.
What you'll do
•
Raise photorealism. Improve materials, lighting, terrain, water, and atmosphere in rendered aerial video, and match camera and sensor effects to real footage. •
Apply generative image and video models. Use models such as Stable Diffusion and SDXL with ControlNet, Stable Video Diffusion, Wan, or Cosmos to enhance rendered footage, with conditioning and automated checks that keep labels accurate. •
Fine-tune on in-domain data. Adapt generative models to specific sensors, environments, and conditions while keeping results temporally consistent. •
Broaden coverage. Extend scene generation to new environments, platforms, object types, and conditions. •
Build the asset library. Source, license-check, and maintain 3D assets with correct scale and materials. •
Measure realism quantitatively. Run sim-to-real experiments and track distribution metrics, not just visual impressions. •
Run at scale. Operate rendering on GPU infrastructure with reproducible, well-documented outputs. What you bring •
2+ years building synthetic data, rendering, or simulation pipelines for computer vision, with at least one pipeline whose output trained a model that went into use. •
Hands-on generative image and video experience. You have run and adapted Stable Diffusion family models (SD 1.5, SDXL, or SD3) and at least one video diffusion model (Stable Video Diffusion, AnimateDiff, Wan, Cosmos, or similar). You know ControlNet or comparable conditioning, and image-to-image and video-to-video workflows. You can show work, not just tutorials. •
3D and rendering fundamentals. You have strong NVIDIA Omniverse experience. You understand PBR materials, path tracing, lighting, and camera models. •
Strong Python engineering. You use NumPy, OpenCV, and PyTorch, write clean, testable code, and work comfortably with Docker, Linux, and GPU nodes.
Raise photorealism. Improve materials, lighting, terrain, water, and atmosphere in rendered aerial video, and match camera and sensor effects to real footage. •
Apply generative image and video models. Use models such as Stable Diffusion and SDXL with ControlNet, Stable Video Diffusion, Wan, or Cosmos to enhance rendered footage, with conditioning and automated checks that keep labels accurate. •
Fine-tune on in-domain data. Adapt generative models to specific sensors, environments, and conditions while keeping results temporally consistent. •
Broaden coverage. Extend scene generation to new environments, platforms, object types, and conditions. •
Build the asset library. Source, license-check, and maintain 3D assets with correct scale and materials. •
Measure realism quantitatively. Run sim-to-real experiments and track distribution metrics, not just visual impressions. •
Run at scale. Operate rendering on GPU infrastructure with reproducible, well-documented outputs. What you bring •
2+ years building synthetic data, rendering, or simulation pipelines for computer vision, with at least one pipeline whose output trained a model that went into use. •
Hands-on generative image and video experience. You have run and adapted Stable Diffusion family models (SD 1.5, SDXL, or SD3) and at least one video diffusion model (Stable Video Diffusion, AnimateDiff, Wan, Cosmos, or similar). You know ControlNet or comparable conditioning, and image-to-image and video-to-video workflows. You can show work, not just tutorials. •
3D and rendering fundamentals. You have strong NVIDIA Omniverse experience. You understand PBR materials, path tracing, lighting, and camera models. •
Strong Python engineering. You use NumPy, OpenCV, and PyTorch, write clean, testable code, and work comfortably with Docker, Linux, and GPU nodes.