Join us
Build the future of multimodal intelligence.
Work on the image, video, audio and lipsync stack we train, serve and run ourselves.
How we work.
One team does the research and runs what it builds.
One stack, end to end
We optimise the whole path ourselves: training, serving and the hardware underneath it.
Measured on real work
A change counts when it raises accepted outputs per server-hour on real workloads, not only on a benchmark.
Research that ships
What works goes into the API, the custom services and the compute we run for customers.
Open positions.
We hire for the four areas below. If your work fits one of them, apply with the form at the end of this page.
Diffusion training and fine-tuning
Train and fine-tune the image and video models behind the product, against the checks that decide whether an output is accepted.
Inference optimization
Make every sampler step cheaper without losing quality: precision, kernels and scheduling on new GPUs.
Serving stack
Batching, scheduling and caching that keep cost and latency predictable under real traffic.
Evaluation and measurement
The checks and measurements behind every claim: what to measure, how, and how much evidence is enough.
Apply.
Tell us what you work on and what you would want to measure first. Links to papers, code or demos help us most.