Topaz Sharpen AI

About Topaz Sharpen AI

Conventional sharpening does one thing: it raises contrast at edges. That makes an image look crisper from a distance and produces halos, crunchy noise and a plasticky texture the moment it is examined closely. This tool takes a different route, using trained models that were shown pairs of soft and sharp frames and learned to reconstruct the detail that the blur removed rather than to exaggerate what survived.

The important step is choosing the right model, because the three problems it addresses are not the same problem. Motion blur comes from a subject or camera moving during exposure and needs directional reconstruction. A focus miss puts the plane of sharpness in the wrong place and needs a different correction. General softness from a weak lens, diffraction at a small aperture or a slightly misaligned sensor is a third case again. The application can pick automatically, and manual selection is usually better.

Everything works on a masked region as well as the whole frame. Sharpening an entire portrait raises noise in the background for no benefit, so the brush restricts the effect to eyes, hair and edges of clothing and leaves the rest untouched. A split preview at full magnification shows the before and after side by side while parameters are adjusted.

It runs standalone on a folder of files or as a plugin inside a host editor, returning a layer rather than a flattened result. Batch processing applies one model and one set of parameters across a whole shoot, which is the sensible way to deal with a set of frames that all share the same lens softness.

What is included in this build

• Separate trained models for motion blur, focus miss and general lens softness
• Detail reconstruction rather than edge contrast enhancement, avoiding halos
• Automatic model selection with manual override when the automatic guess is wrong
• Masking brush so sharpening applies only where it helps
• Split preview at full magnification while parameters are adjusted
• Batch processing across a folder with one model and one parameter set
• Standalone application and host editor plugin, returning a layer rather than a flattened frame
• GPU acceleration with a CPU fallback on machines without a suitable card

Changed in version 4.1.0

• Improved motion blur model on frames with mixed subject and camera movement
• Faster processing on current generation GPUs
• Masking brush edge handling refined around hair and fine detail
• Batch queue now resumes after an interruption
• Reduced memory use on very high resolution frames

Installation walkthrough

These steps assume a clean machine. If a previous release of the same application is present, deal with that first, since an overlapping install is the most common cause of a setup that stops halfway.

1. Install the current graphics driver from the GPU vendor before first launch.
2. Disable real-time protection until installation has finished.
3. Extract the archive to a short path and run setup as administrator.
4. Let the model files finish installing before opening the application.
5. Point the plugin path at the host editor if the plugin is wanted as well.

System requirements and file details

Frequently asked about this title