About this release
Topaz DeNoise AI does one thing and does it better than the noise reduction built into a general editor. The problem with traditional noise reduction is that it cannot tell noise from fine detail, so turning it up to clean a grainy image also smears away the texture you wanted to keep, and the result looks waxy. This tool uses a model trained on the difference, so it removes the noise and leaves the detail.
The practical use is rescuing images that a general workflow gives up on, a high ISO frame shot in poor light, an underexposed capture pushed hard in editing, or an old image from a noisy sensor. It offers several processing modes for different noise characters, and it can recover and sharpen genuine detail at the same time as it cleans, so the output often looks better than a clean low ISO shot would have.
It runs as a standalone application for a quick single image pass and as a plugin into a larger editor so the denoise step fits into a bigger workflow. Processing is accelerated on the graphics card, and a batch mode runs a whole folder of noisy frames through in one go. This release is the full application unlocked from first launch.
What it does
- Trained model that separates real detail from sensor noise
- Several processing modes for different noise characters
- Detail recovery and sharpening alongside the noise removal
- Before and after split view for judging the result
- GPU accelerated processing for speed
- Batch mode for a whole folder of noisy frames
- Raw file support for cleaning before the rest of the edit
- Runs standalone or as a plugin in a larger editor
- Non destructive workflow that keeps the original untouched
- Adjustable strength so cleaning never oversmooths
Changes in this version
- Faster processing on current generation graphics cards
- Improved detail retention at very high noise levels
- Reworked mode selection with clearer previews
- Better raw file handling before demosaicing
- Reduced memory use on high resolution files
What the package contains
- Topaz DeNoise AI standalone installer
- Plugin build for a larger host editor
- Bundled processing models
- Readme with the install and plugin setup steps
- Notes on batch processing a folder
Installation notes
Follow these in order. Most reports of a failed install come from skipping a step rather than from the package itself.
- Disconnect from the network before you begin.
- Pause real time protection for the duration.
- Extract the archive to a folder you control.
- Run the installer as administrator and finish the wizard.
- For plugin use, point the installer at your host editor's plugin folder.
- Apply the included unlock step, launch once, then reconnect.
System requirements
| Operating system | Windows 10 version 22H2 or Windows 11 |
| Processor | Intel or AMD 64-bit with four or more cores |
| Memory | 8 GB minimum, 16 GB recommended |
| Graphics | GPU with 4 GB VRAM for accelerated processing |
| Storage | 4 GB free during install |
| Display | 1920 by 1080 or larger |
Things worth knowing
- The processing wants a capable GPU, without one it falls back to the processor and slows right down.
- Denoise before heavy sharpening in your workflow, cleaning after sharpening fights itself.
- Push the strength too far and even this smooths detail, judge it in the split view before committing.
Questions about this title
How is it different from the noise reduction in my editor?
Does it work as a plugin?
Can it clean a whole shoot?
Does it need a GPU?
What people are saying
Rescued a whole dim reception set shot at 12800. Cleaned the noise and kept the detail. Remarkable.
Use it as the first step on my night frames. Nothing built into an editor comes close on grain like this.
Ran a folder of noisy frames overnight. Grouped them by ISO first like the notes say. Consistent results.
Listing information
This entry is catalogued under Photo Editing and carries the tags below. Language coverage is English, and the supported platform list is Windows 10, Windows 11. Version 3.7.2 is the current catalogued build, last checked 1 week ago, and the entry has been in the index since it was added 2 years ago.