Overview
Conventional sharpening increases local contrast at edges, which is why an over-sharpened photograph has bright fringes around every dark line and a gritty texture in areas that should be smooth. Sharpen AI works differently. It uses trained models to estimate what caused the softness and reconstructs the detail accordingly, which produces a result without the halo signature.
The important design decision is that the three common causes are handled by separate models rather than one slider. Motion blur from a moving subject or a shaky hand is a directional problem. A focus miss is a depth problem where the plane of sharpness landed in the wrong place. General softness from a weak lens, a diffraction limited aperture or a heavy anti-aliasing filter is a different problem again. Each gets its own model, and the application will suggest which one applies.
It runs as a standalone application and as a plug-in for the common editors, so it can sit at the end of an existing retouching chain and return a layer rather than replacing the workflow. Masking limits the effect to the parts of the frame that need it, which matters because sharpening a background that is meant to be out of focus is exactly the wrong result. Batch processing runs a folder through one recipe.
What it does well
Cause specific models
Separate models for motion blur, focus miss and general softness, with an automatic suggestion for which one the image needs.
No halo signature
Detail is reconstructed rather than edge contrast being raised, so the result does not carry the bright fringing conventional sharpening leaves.
Masking
Brush or select the region to sharpen so out of focus backgrounds and smooth skin stay as they are.
Plug-in and standalone
Runs on its own or inside the common editors, returning a new layer at the end of an existing retouching chain.
Batch processing
Run a folder of images through the same recipe with per-image model selection kept where it was chosen.
Changes in this build
- New model with better texture retention on fine hair and fabric.
- Faster processing and lower VRAM use on current GPUs.
- Automatic model suggestion improved on mixed cause images.
- Comparison view rebuilt with a synchronised split and a zoom lock.
- Batch queue now keeps per-image settings rather than applying one global recipe.
What is in the package
- Sharpen AI installer, 64-bit
- Complete model weight set
- Plug-in components for the common host editors
- Batch queue and preset library
- Sample images for model comparison
System requirements
| Processor | Intel or AMD 64-bit, quad core or better |
| Memory | 8 GB minimum, 16 GB for high resolution files |
| Graphics | GPU with 4 GB VRAM recommended, 2 GB minimum |
| Storage | 6 GB free |
| Display | 1920 x 1080 minimum, colour accurate panel recommended |
| System | Windows 10 22H2 or Windows 11, 64-bit |
Installation
- Extract the archive to a local drive.
- Close the host editor if you want the plug-in registered cleanly.
- Run the installer and let the model weights unpack fully.
- Apply the included configuration step before first launch.
- Open a sample image and run a preview to confirm GPU processing is active.
Before you start
Applying it to an already sharpened file compounds the effect, work from the unsharpened original.
The models are large, the install takes longer than the application size suggests.
Masking is not optional on portraits, an unmasked pass will sharpen skin texture you did not want.
Questions about this title
Are the models bundled?
Yes, the full weight set installs with the application and nothing is fetched later.
Does it work as a plug-in?
Yes, for the common host editors, returning the result as a new layer.
Can it fix motion blur?
Within limits. Mild directional blur recovers well, severe blur does not.
Does it need a GPU?
It runs on CPU but slowly. A GPU with 4 GB is the practical recommendation.
About this listing
This entry was checked on a clean install of Windows 10 / 11 (64-bit) before it was published, and it is rechecked whenever the package is rebuilt. The figures on this page come from the site index rather than from the publisher, so the download count is what people here have actually pulled.
If a mirror stops answering, use the contact form and it gets replaced. Reports about a specific title are handled faster than general messages because they name the file, the mirror and the point at which the transfer stopped.
Requests for a different version, a different language build or an older release go on the requests page. Older versions of a title are usually still held even when only the current one is listed.