Topaz Video AI 1.7.1 for macOS

How it gets your old video looking right

Topaz Video AI is desktop software that sharpens, upscales, and steadies your video using machine learning. Think of it as a dedicated cleanup crew for footage that’s seen better days — old DVD rips, shaky DSLR clips, grainy game captures, and even archival family films.

The workflow is pretty straightforward once you get the hang of it. Here’s how most people approach a project.

1. Load your source clip. Drag in whatever you’re working with — standard definition home movies, noisy low-light footage, or HD stock shots you need at 4K or even 8K.
2. Pick your fix. The app lets you target specific problems. Remove grain and compression artifacts, generate new frames to smooth out motion or change frame rate, or stabilize hand-held wobble.
3. Set the output size. Tell it where you’re headed. Upscaling from SD to HD is a common job, and it handles the pixel math to get there without the blocky mess you’d get from a simple resize.
4. Let it run. Processing takes a while — it’s rebuilding frames from scratch. Grab a coffee; this is normal.

It’s built for folks who work with their own footage — editors rescuing old commercials, filmmakers prepping music videos, or YouTubers cleaning up game recordings. It runs on Apple Silicon and Intel Macs with macOS 11 or later.

Things don’t always go smooth the first time. Here are the snags I hear about most and what to do when you hit them.

Fixes for garbled exports and stuck previews

The export takes forever and seems stuck

This is the number one worry. Topaz Video AI isn’t a quick render — it’s generating new pixels frame by frame. A ten-minute 1080p clip upscaled to 4K can easily run for hours. Before you force-quit, check two things. First, look at the preview panel — if the little progress bar is still crawling forward, it’s working. Second, hover over the time-remaining estimate. It recalculates constantly during the first few frames, so early numbers are often wildly wrong. Let it settle for five minutes before you decide it’s broken.

Patch.command throws an error on newer macOS

If you’re on a recent version of macOS, you might double-click that Patch.command file and get a security warning or a generic error instead of a password prompt. The fix is simple: right-click the file and choose Open from the menu, then confirm you want to run it. That bypasses Gatekeeper’s blanket block. Once the Terminal window pops up, type your Mac password — the characters won’t show on screen, which is normal — and press Return.

The output looks plasticky or over-smoothed

This usually means you’ve cranked a slider too far, or you picked a model that’s too aggressive for your source. Back off the denoise strength first. If faces in the frame look waxy, the AI is scrubbing away real detail along with the noise. Try switching to a gentler model — Proteus handles most natural video well, while Artemis is tuned for noisy digital footage. A quick ten-second test export saves you from discovering the problem after an overnight render.

Stabilization makes the frame wobble at the edges

The stabilization model crops in to keep the frame steady, and when the original shake is severe, that crop gets deep. If the result bounces at the edges, reduce the stabilization strength or switch off the Auto-crop toggle so you can set the crop manually. Sometimes leaving a thin black border that you trim later in your editor is cleaner than letting the AI guess the framing.

Version 1.7.1 makes preview scrubbing feel instant

The app used to be called Video Enhance AI, and the rename to Topaz Video AI came with more than a new logo. Version 1.7.1 brings a few quality-of-life improvements that change what you notice day to day.

• Faster preview generation. Earlier builds made you wait through a full model load just to see a single test frame. The preview system now uses shorter sample clips by default, so you can compare models side by side without a ten-minute coffee break between each one.
• Better Apple Silicon handling. The app now detects M-series chips and assigns processing threads more sensibly. On an M1 or M2 Mac, you’ll see fewer hangs during model switching — something that used to trip up the older Intel-optimized code paths.
• Refined frame interpolation. Chronos, the frame-rate model, got tweaks that reduce the ghosting around fast-moving subjects. If you convert 24 fps to 60 fps, arms and legs in motion leave fewer smeary trails than they did in the 1.5-era builds.

What I’d tell you on day one

After installing the pkg and running the patch, resist the urge to throw your longest, most precious project at it immediately. Spend your first hour on test clips instead.

• Create a short test library. Cut three ten-second samples from different kinds of footage you actually work with — one noisy, one soft-focus, one shaky. Run each through two different models at a modest upscale target before you commit to a full project. You’ll learn how the sliders behave on your material, not the demo reel.
• Check your scratch disk space. The app writes temporary frames during processing, and a 4K output can eat 50 GB of free space on a big job. Point the cache folder to an external drive with room to breathe, or clear it out after each completed export.
• Set the output format early. The default container might not match what your editor expects. If you’re dropping the result into DaVinci Resolve or Final Cut, pick ProRes or DNxHR rather than h.264 — you’ll save yourself a transcoding headache later.

Stacking too many models amplifies artifacts

Everyone makes these. I did too.

Throwing every fix at once. It’s tempting to check denoise, sharpen, upscale, and stabilize all in one pass. The models stack, and each one introduces tiny artifacts that the next one amplifies. Pick the worst problem first. If the clip is noisy, clean it up. If it’s soft, sharpen it. Run separate passes and compare the results — the extra time is worth it.

Upscaling something that’s already been badly compressed. A grainy 480p clip that’s been through YouTube’s compression wringer twice won’t magically become crisp 4K. The AI can only work with the data that’s actually in the frame. If the source is a blocky mess, lower your expectations and aim for a clean 720p or 1080p instead. You’ll get something watchable rather than a high-resolution record of compression artifacts.

Ignoring the preview entirely. The little preview window isn’t a suggestion — it’s an accurate sample of what the whole export will look like. Zoom in on the preview and check faces, text, and high-contrast edges. If those look wrong, the full render will look wrong too, just larger.

Letting the Mac go to sleep mid-job. By default, macOS might nap after a while. An interrupted render often can’t be resumed — you lose the hours already spent. Set System Settings > Battery > Prevent automatic sleeping before a long export, or use a utility like Amphetamine to keep things awake.

• macOS 11 or newer
• Works on Apple Silicon and Intel 64-bit Macs