Topaz Video AI Pro

About Topaz Video AI Pro

Topaz Video AI does the jobs conventional filters cannot. Upscaling standard definition archive footage to high definition, cleaning heavy sensor noise out of low light material, deinterlacing broadcast tape properly, and generating intermediate frames for slow motion are all handled by trained models rather than by interpolation kernels, and the difference on difficult material is large.

There is no single model, and picking the right one matters. Separate models are trained for compressed footage, for clean digital sources, for interlaced broadcast material and for heavily degraded archive film. The preview compares models side by side on a short section before committing an hour of processing to the wrong choice.

Frame interpolation is the other main use. Generating intermediate frames produces smooth slow motion from footage shot at a normal frame rate, or converts between frame rates without the judder a simple duplication pass leaves behind. Motion is estimated rather than blended, so moving edges stay sharp.

Processing is GPU bound and honest about it. The queue runs unattended across a list of files, exports to standard intermediate and delivery codecs, and reports realistic time estimates. Trimming to the section that actually needs work before processing saves hours on a long source.

What is included in this build

• Machine learning upscaling up to 4K and beyond from low resolution sources
• Separate trained models for compressed, clean, interlaced and archive material
• Side by side model comparison on a short preview section before committing
• Denoise and detail recovery on low light and high ISO footage
• Proper deinterlacing of broadcast and tape sources
• Frame interpolation for slow motion and frame rate conversion
• Unattended batch queue across a list of source files
• Export to standard intermediate and delivery codecs

Changed in version 7.1.6

• New model added for heavily compressed source material
• Faster processing on current generation GPUs
• Improved stability on very long source files
• Preview comparison interface reorganised
• More accurate processing time estimates in the queue

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 GPU driver before starting, since processing is entirely GPU bound.
2. Disable real-time protection until the install has finished.
3. Extract the archive to a short path and run setup as administrator.
4. Let the bundled models finish unpacking before launching.
5. Test a short clip first to confirm the model choice before queueing a long source.

System requirements and file details

Frequently asked about this title