We find the signal
inside the noise.
Denoise Labs builds intelligent systems that make complex media clearer, more useful, and more human.
Discover our approachClear images from noisy, incomplete data.
Denoise Labs is the studio of Jasper Schoormans, a freelance AI developer with a PhD in medical imaging from the University of Amsterdam, where he reconstructed sharper MRI from sparse, noisy measurements. Today we bring that same idea to generative AI, from first experiment to production.
We work with companies and product teams that need custom image AI built and shipped, tuned to their own data instead of off-the-shelf tools.
Tell us about your data and what you want to build.
-
01
Generative AI
We build custom diffusion pipelines, ControlNets and LoRAs around your data and use case.
-
02
Image generation
We design generative image workflows, from virtual staging of empty rooms to photo enhancement.
-
03
Image reconstruction
We denoise, restore and reconstruct images when the data is noisy or incomplete.
-
04
Data science & research
We explore your data, prototype models and turn open questions into tested answers.
-
05
Deployment
We ship models to production as inference, APIs and pipelines your team can actually run.
From MRI reconstruction to generative imaging.
-
Interesting Pixels · 2025
Custom ControlNets for virtual staging ↗
A custom ControlNet trained to stage empty rooms from RGB input, keeping windows, doors and room structure intact.
-
Interesting Pixels · 2025
IC-LoRA for panorama virtual staging ↗
A LoRA trained on 3D renders with and without furniture, to stage empty panorama rooms automatically.
-
PhD thesis · University of Amsterdam · 2020
Innovations in sparse MRI: in search of a free lunch ↗
Undersampling strategies with advanced reconstruction methods to make MRI acquisitions faster and more efficient.
-
Physics in Medicine & Biology
Compressed sensing MRI with variable density averaging ↗
Undersampled k-space with variable-density averaging gave better image quality than full sampling in the same scan time at low SNR.