Topaz Video AI in 2026: Subscription-Only, Adobe-Owned, and Still the Best Upscaler

The version of Topaz Video AI that a lot of creators bought for a flat $199 no longer exists. On October 3, 2025, Topaz Labs stopped selling perpetual licenses for its whole lineup. Version 7.1.5 was the last one you could buy outright, and it will not receive new models. Nine months later, on June 25, 2026, Adobe announced it was acquiring the company. So the tool many of us recommended precisely because you owned it forever, with no subscription, is now a yearly subscription that a $200-billion software giant is about to own.

That is a big enough shift that the old “buy it once, process 1,000 videos, done” pitch is dead. The upscaling engine, though, is still the best one you can run on your own machine. This is the honest 2026 picture: what changed, what it costs now, which models actually matter, and whether it still earns a slot in your workflow.

Two years, three product changes

If you last looked at this software in early 2024, three things happened while you weren’t watching.

First, the rename. “Topaz Video AI” is now just Topaz Video. Its siblings dropped the “AI” too: Gigapixel AI became Topaz Gigapixel, Photo AI became Topaz Photo. Same engines, cleaner names.

Second, the license switch. From October 3, 2025, every product moved to subscription only. Existing perpetual owners keep the last version they paid for, but that copy is frozen in time. New AI models, and there are a lot of them now, require an active subscription. Let your subscription lapse and the app stops opening, which is the part longtime users are angriest about. Digital Production and CG Channel both covered the backlash when it landed.

Third, the acquisition. Adobe entered a definitive agreement to buy Topaz Labs, expected to close in the second half of 2026 pending regulatory approval. Topaz CEO Eric Yang stays on to lead the team, and Adobe says the standalone products will keep selling on topazlabs.com. I wrote a full breakdown of what the Adobe deal actually changes for creators, because the buyout is less about the upscaling app you use today and more about the engine underneath it. More on that below.

What it costs now, in plain numbers

Here is the pricing as it stands in 2026, before the seasonal promotions that Topaz runs constantly. Treat the exact figures as a snapshot, not a contract.

Sources: Topaz’s own Precision Update page for the Studio and Astra figures, and the pricing breakdowns tracked at Nero’s Topaz comparison.

Read the table like an operations person, not a shopper. The old math was simple: $199 up front, then zero. The new math is recurring, and it splits your work into two buckets. Local models run on your GPU for no per-video cost once you are subscribed. The best generative models run in the cloud and burn credits. That second bucket is where a heavy month gets expensive, and it is the number to watch if you process a large archive.

In 20-plus years running IT operations, I have watched this exact transition play out across dozens of vendors. Perpetual to subscription is rarely about the sticker price. It is about who controls the off switch. With a perpetual license, a tool you bought in 2021 still opens in 2026. With a subscription, access is a monthly decision the vendor makes, not you. That is not automatically bad, but it changes how you should think about depending on the tool for client work you have already been paid for.

What the engine still does better than the free options

None of the business drama changes the core reason people reach for this software: it reconstructs detail that a cheaper scaler cannot invent. A few capabilities still stand out for creators.

Upscaling old footage. The headline job. Feed it a 1080p tutorial from 2020 and it rebuilds toward 4K by predicting texture and edge detail from patterns it learned across millions of frames, rather than stretching pixels. For a channel sitting on a back catalog that still pulls organic views, that turns an archive into new uploads instead of dead files.

Noise reduction on low-light footage. The denoise models compare frames over time to separate real detail from grain. That grainy restaurant interview or the murky wedding reception under string lights becomes deliverable instead of an apology.

Frame interpolation. It generates in-between frames to lift 24 or 30fps up to 60fps, or to build convincing slow motion from standard-speed clips. Simple camera moves and product shots interpolate cleanly; fast, chaotic motion still produces artifacts.

Stabilization that keeps your frame. Traditional stabilizers crop in to absorb shake. Topaz analyzes the full sequence and often holds more of your original composition, which is the difference between usable handheld B-roll and footage that makes viewers queasy.

If your work is mostly modern, well-shot 4K already, you may never need any of this. The value is concentrated in restoration, salvage, and archive revival, not in polishing footage that is already clean.

The models that matter in 2026

The old model names some guides still list (Artemis, Gaia, Nyx) are legacy entries. The 2026 lineup is bigger and splits into two philosophies, a distinction Topaz introduced with its March 2026 Precision Update.

Precision models enhance the detail that is genuinely in your footage. This family includes Proteus for general live action, Iris, Rhea and Rhea XL for texture-heavy 4x upscales, and Gaia 2, which is tuned for animation and motion graphics. These are the safe, faithful choices when the source is decent and you want a natural result.

Generative models infer detail that was never captured, building structure and texture from learned context. Wonder 3 sits here, and so does the one everybody talks about: Starlight, a diffusion-based model that is the current gold standard for reviving VHS-era and 360p footage with temporal consistency nothing else matches. Starlight comes in several flavors: Starlight Fast 2 runs locally, Starlight HQ handles higher-resolution inputs, and Starlight Precise 2.5 is cloud-only through Astra and is aimed at cleaning up AI-generated video.

The practical rule: precision models for real footage you want to stay honest, generative models when the source is so degraded that faithful enhancement has nothing to work with and you are willing to let the AI invent. Preview a short section before committing an overnight render, because the wrong model on the wrong clip wastes hours. And resist maxing every slider. Subtle enhancement reads as professional; obvious AI processing reads as, well, obvious.

NeuroStream is the real reason Adobe paid

Here is the part that reframes the whole acquisition. Adobe did not primarily buy an upscaling app. It bought NeuroStream, Topaz’s engine for running large, complex AI models locally on consumer hardware. Topaz says it cuts the VRAM these models demand by up to 95 percent, which is what lets heavy video models run on an ordinary desktop GPU instead of a cloud server or a workstation.

That directly rewrites the old hardware advice. Guides from a couple of years ago told you to expect a hard 8GB VRAM floor for 4K work. NeuroStream is exactly the technology chipping away at that floor, and it is why processing that used to be cloud-only is increasingly landing on local machines. It is also why Adobe wanted the company: on-device AI that runs in Photoshop, Lightroom, and Premiere without a round trip to a server. Adobe president David Wadhwani framed the deal around creators “mixing captured and generated images and video,” and the plan is to weave the tech into Firefly and the Creative Cloud apps. If you already live in Adobe’s Creative Cloud after its June 2026 AI update, this restoration quality is coming to you whether or not you ever open the standalone Topaz app.

The local-processing angle matters beyond speed. Your footage never leaves your machine, which is a genuine data-governance point for anyone handling client or confidential material. Cloud upscalers upload your source; local ones do not. For a fractional COO reviewing a creator’s tool stack, “does this send client footage to someone else’s server” is a real question, and the local models answer it cleanly.

How I would actually use it on an archive

The workflow that pays for the subscription is systematic, not random.

Start with analytics, not nostalgia. Pull the videos that still earn organic traffic but look dated next to your current output. You already know there is demand; you are just raising the resolution to meet it. Process similar content with the same model and settings so a series keeps a consistent look. Export enhanced masters, then build genuinely new edits on top of them, updated graphics, cleaner audio, an extra segment, so the re-release feels intentional rather than a lazy re-upload. This is the same archive-revival logic that makes a faceless YouTube channel’s back catalog keep compounding.

For client work, build render time into the timeline and the quote. If you promise three days of editing, an overnight Topaz pass fits inside that; a same-day news turnaround does not. Keep a naming convention that survives a deadline scramble: Client_Interview_Original.mov next to Client_Interview_Topaz4K.mov. And watch the cloud-credit meter on big jobs, because that is the line item that turns a predictable subscription into a surprise.

Where it fits, and where it does not

Free and built-in options. DaVinci Resolve 21 includes Super Scale upsampling at no cost, and it is fine for light upscaling if you already edit there. Premiere’s own AI upscaling runs inside a Creative Cloud plan you may already pay for. HandBrake handles format conversion and basic cleanup for free but invents no new detail. For occasional, undemanding jobs, one of these is the honest answer, and there is no reason to add another subscription.

When Topaz still wins. Heavy restoration, VHS and camcorder rescue, salvaging low-light shoots, and any archive large enough that quality differences compound across dozens of videos. That is where Starlight and the precision models pull far enough ahead to justify the recurring cost. If AI generation is more your lane than restoration, our complete 2026 guide to AI video tools covers the create-from-scratch side.

When to walk. If your footage is already clean and modern, if you need real-time turnaround, or if you simply refuse to rent software you used to own. That last reason is legitimate. Plenty of creators are keeping their frozen 7.1.5 perpetual copy for as long as it runs and pairing it with a free scaler for everything else.

The call for 2026

Topaz Video still produces the best local upscaling and restoration a creator can run, and NeuroStream is about to make that quality both faster and far more widely available, inside Adobe’s apps as well as its own. What you are weighing is no longer a one-time $199 versus the competition. It is an annual subscription, a cloud-credit meter, and a company mid-acquisition, against an engine that genuinely has no equal for reviving old footage.

If restoration is a real and recurring part of your work, the subscription earns its keep, and the Adobe deal probably improves the tool rather than harming it. If you only touch upscaling now and then, keep your money, lean on Resolve or Premiere, and revisit once the acquisition closes and the pricing settles. Either way, decide based on how the tool is priced and owned today, not on the flat-fee reputation it earned in a version you can no longer buy.