Founding ML Engineer — Diffusion
Own the diffusion models that turn a flat product shot into photoreal on-model imagery — from research to production inference.
About the role
Piktor generates photoreal on-model imagery from a single flat product photo. The model is the product. As our founding ML engineer, you'll own the diffusion pipeline end to end — training, fine-tuning, evaluation, and the low-latency inference that returns a 4K on-model render in under 15 seconds.
This is a deep, high-ownership role on a small team. You'll set the technical direction for image quality — garment fidelity, fabric drape, identity control, clean masks — and turn research into infrastructure that hundreds of brands rely on every day.
What you'll do
- Train, fine-tune and evaluate diffusion models for garment-accurate, photoreal on-model generation.
- Build and own the inference stack: sub-15-second, 4K-quality generation at catalog scale.
- Push image quality on the details that matter — fabric drape, fit, lighting, identity consistency and clean masks.
- Design evaluation harnesses and human-in-the-loop review so quality is measured, not guessed.
- Optimize cost and latency across GPUs, batching and caching without compromising fidelity.
- Partner with product and design to translate art-direction controls (pose, light, model) into model parameters.
What we're looking for
- Strong applied experience with diffusion models and modern generative image pipelines (latent diffusion, ControlNet-style conditioning, LoRA / fine-tuning).
- Fluent in PyTorch and comfortable owning training and inference end to end.
- A track record of shipping ML to production — not just notebooks and papers.
- Solid grasp of GPU performance: latency, throughput, quantization and memory trade-offs.
- Pragmatic research taste — you know which quality problems are worth solving and which aren't.
- Comfortable operating with high autonomy in an early-stage, fast-moving team.
Nice to have
- Published work in generative vision, or a portfolio of relevant open-source projects.
- Experience with virtual try-on, human / garment modeling or 3D.
- Familiarity with serving frameworks and inference optimization (TensorRT, Triton, compiled kernels).
- A genuine interest in fashion, imaging or e-commerce.
How to apply
Email hello@piktor.co with a short note on the most interesting generative-imagery problem you've shipped, plus links to papers, code or work we can look at. No cover letter needed — show us something real.
Not quite the right role?
Browse every open position, or introduce yourself — we love meeting people ahead of the need.