Cloud and desktop upscalers solve the same problem with opposite cost structures. One charges per job and needs no hardware; the other charges a subscription and wants a strong GPU. This guide compares the two honestly — no universal winner, just the factors that decide which fits your situation.
What changed with Topaz pricing
Topaz Labs announced in September 2025 that it would stop selling perpetual licenses for its products, and perpetual sales ended fully in early October 2025. Topaz Video (formerly Video AI) is now sold as a subscription: $59 per month on a month-to-month plan, or $299 per year on an annual plan. The software runs on Windows and macOS only.
Existing perpetual licenses from before the cutoff still work for the versions they cover, but new users cannot buy one.
Pricing details change; always check the official Topaz Labs page before making a purchase decision. The figures above are our understanding at the time of writing, based on public announcements.
Cloud: SeedVR2.net
SeedVR2.net runs the upscaler in the browser. You upload a clip, pick a model and target resolution, and download the result — no GPU, no install, no API wiring.
How it charges:
- Credits per job: images cost 10 (2K), 16 (4K), or 25 (8K) credits
- Video bills per second with a 5-second minimum; the standard model is 9 credits per second at 720p/1080p, 33 at 4K
- New accounts get a free first 2K image and a free first video run (standard model, 720p, up to 5 seconds, verified email)
- Credit packs start at $29.99 for 1,500 credits; current pricing is on the pricing page
The tradeoff is straightforward: files are uploaded for cloud processing, and you pay per job. For occasional use, that structure is hard to beat — a one-off 30-second 4K clip on the standard model is 990 credits, with no subscription attached.
Desktop: Topaz
A desktop suite processes files locally on your own hardware. Topaz Video adds enhancement, stabilization, and frame interpolation alongside upscaling, and local processing means files can stay on your machine.
The costs sit in three places:
- Hardware: a capable GPU with enough VRAM is effectively required for reasonable render times
- Subscription: $59 per month or $299 per year, ongoing
- Platform: Windows and macOS only — Linux is not supported
The subscription makes most sense at high volume: if you upscale hours of footage weekly, the fixed cost amortizes across hundreds of jobs. At low volume, the same $299 buys thousands of cloud credits instead.
Cost comparison at different volumes
- One or two jobs a month: cloud wins almost by default. You pay a few dollars' worth of credits and nothing else
- Weekly creator workload: both are viable. Compare a month of credits against the subscription, and factor in whether you already own the GPU
- Professional studio, daily batches: the desktop subscription plus local batch processing is the conventional choice — but verify current terms before committing
The site keeps a longer four-way comparison of SeedVR2.net, Topaz, local ComfyUI, and API providers if you want the full picture.
Privacy and file handling
The sharpest difference: cloud processing means your source file is uploaded to a hosted service, while desktop processing can stay entirely on your machine. For client footage with confidentiality requirements, local processing is often mandatory — that constraint alone can decide the question regardless of price.
How to decide
Run this checklist:
- How many jobs do you do per month? Under ten favors credits
- Do you already own a capable GPU? If not, a desktop suite starts with a hardware purchase
- Does the footage need to stay local? If yes, desktop is the only option
- Are you on Linux? Topaz is not available there; cloud works in any browser
- Do you need extras like stabilization and interpolation? That is Topaz territory
Neither approach is universally better. SeedVR2.net fits people who want a finished result in the browser without hardware or a subscription — the video upscaler quotes the exact credit cost before it runs. Topaz fits people with a workstation and high local volume. Compare on your real numbers, not the marketing pages.
What a logged-out ChatGPT did with this exact choice on 9 October 2026
Asked for the best AI video upscaler in 2026 for old footage, a ChatGPT session with no account and no saved chats named Topaz Video first and attached four product homepages: Topaz, AVCLabs Video Enhancer AI, Aiarty, and VideoProc. It did not mention this site. That is the useful result. A stranger's assistant still treats "old footage, more control" as a desktop-suite question, and it answers with vendor homepages rather than a benchmark.
The same session, asked whether any free upscaler works without login, attached Video2X and pages that advertise no sign-up. It did not attach a desktop subscription. So the split in this article is the split ChatGPT already makes: long, private, or heavily controlled work goes to an installed app; a quick "can this even be fixed" question goes to whatever page promises a free browser pass.
Use that split on your own file. If the footage cannot leave the machine, stop here and install the desktop app ChatGPT cited. If you can upload five seconds, the hosted sample is 720p, standard model, one run per verified account. It is not 4K, and it is not a substitute for Topaz's local filter stack. Standard credit rates after the sample are 9 per second at 720p and 1080p, 17 at 2K, and 33 at 4K, with a five-second minimum. Thirty seconds at 4K is 990 credits. The calculator is the same table the tool shows before submit.
One limit covers both columns. Desktop and cloud models reconstruct. They do not recover texture the tape or the encoder never stored. If the five-second cloud sample looks invented, a longer Topaz render of the same broken shot will spend more time on the same missing detail. Keep the capture. Export a copy. Write the output size you accepted next to the file name so the next session does not re-buy the same test.