Cloud workers
Planned- Input
- Image and video batches
- Output
- Mattes, layers, enhanced frames
- For
- E-commerce and media pipelines
Two research engines for the details a media workflow depends on: the difficult edge, and the difference between recovered detail and a guess.
In researchRequest access
Neural matting and retouching
Separate subjects, garments and accessories from any background without a green screen. Then remove wrinkles and blemishes while keeping the character of the material.
ExtractRefineRecompose
For fashion e-commerce, creative agencies and finishing houses.
PhotoEdges
Needs reviewForensic super-resolution
Enhance degraded CCTV and low-bitrate frames, with a pixel-level uncertainty map that tells an operator where the model may be synthesising plausible detail instead of recovering it.
For security auditing teams reviewing degraded footage.
A sharper image invites more trust than it has earned. The gate is designed to push back.
A map, not a verdict.
The gate produces a variance map across the frame. High-variance regions are where the source signal no longer supports the output.
Human review by default.
Flagged regions are surfaced to the operator instead of being smoothed over. The model never decides what counts as evidence.
Calibration comes first.
The estimator and its thresholds need calibration on representative footage before any output is used in a review.
Shown in amber, always labelled.
Amber is reserved for uncertainty across our products, and it always comes with words, never colour alone.
Local CLI
Planned- Input
- Studio files on disk
- Output
- Layers and uncertainty maps
- For
- Post-production and finishing
Tell us what you need to protect, extract or inspect. We'll tell you honestly whether our engines can help yet.