Why there is no AI model here yet
A learned deblur model reconstructs edges from what it learned on other photos, which is powerful and also the way false detail gets in. This site ships one only when it measurably beats the classical pass on synthetic smears, produces no more artifacts on real photos, runs in an acceptable time on a mid-range phone, and has a clean licence chain. Until then the method page lists what is planned and what runs today.
Questions
Is there an AI deblur model on this site?
Not yet. A learned motion-deblur model is planned, to run on your device after a one-time download whose exact size will be shown first. It will only be switched on once it beats the classical pass on our test set and its licence chain is recorded. Until then, this page runs the classical pass and says so on the method chip and in the receipt.
What runs instead?
The classical pass: the blur is measured off the photo, and a deconvolution against that kernel runs in your browser — an iterative pass for short kernels, a frequency-domain pass for long ones — with an artifact guard. It adds no learned detail, but it can create halos, so compare at 100%.
Will my photo be uploaded when the model arrives?
No. The model file comes to this tab; your photo never leaves it. That is the whole design.
When won't deblurring work?
Long smears, heavy noise, mosaic or pixelation, and detail that was never recorded. The measurement says which, before anything runs.
Related
- Unblur an imageDiagnose the blur, then reconstruct it, with the kernel on a dial.
- How it worksWhich engine ran, how each verdict is reached, and the model licence chain.
- What can be reconstructedLearned priors, the GoPro corpus, and reading a result sceptically.
- What a model download costsWhat is fetched, from where, what is kept and how to clear it.
