Topaz Video AI Pro

About this release

Topaz Video AI does one difficult thing very well: it takes footage that was shot at a resolution or frame rate that no longer fits, and rebuilds it to something usable. Old standard definition material, compressed web downloads, interlaced broadcast tape and shaky handheld footage all come out the other side substantially better than any conventional scaler could manage.

The models are trained for specific problems rather than being one general filter. There are separate models for upscaling clean digital footage, for restoring heavily compressed material, for film grain, for interlaced sources and for computer generated content. Choosing the right one matters more than any setting, and the preview compares several side by side on a section of the clip before committing.

Frame interpolation is the second major function. Footage converts between frame rates by generating intermediate frames rather than duplicating, which makes standards conversion clean, and the same engine produces convincing slow motion from material that was never shot for it.

Stabilisation and deinterlacing round out the set. The stabiliser analyses full camera motion and can either lock the frame or smooth it while preserving intentional movement, and the deinterlacer produces progressive frames without the combing artefacts that simple methods leave behind.

What it does

Task specific models

Separate trained models for compression damage, film grain, interlaced sources and clean digital footage.

Side by side preview

Compare several models on a section of the clip before committing to a full render.

Frame interpolation

Generate intermediate frames for standards conversion and convincing slow motion.

Stabilisation

Full motion analysis that can lock the frame or smooth it while keeping intentional movement.

Deinterlacing

Progressive output from broadcast sources without combing artefacts.

Batch processing

Queue a whole folder with the same settings and let it run unattended.

Changed in this version

• New model for heavily compressed web sources.
• Faster processing on current generation graphics cards.
• Improved detail retention on faces at high upscale factors.
• Preview comparison reworked to show more models at once.
• Better memory handling on long clips.

System requirements

Install order

1. Unpack the archive.
2. Update the graphics driver before first launch.
3. Run the installer and let the model files unpack completely.
4. Process a short test section before committing a long clip to a full render.

Worth knowing before you start

File details