Topaz Video AI

Where Topaz Video sits in the video-enhancement market

Topaz Video (long marketed as Topaz Video AI) is the tool I reach for when the question is "can this footage be saved" — archival tape transfers, old family video, low-bitrate web footage, noisy night shots — rather than "can this footage be generated." It is a desktop application for macOS and Windows, also available as a plugin, that applies AI models to upscale, denoise, sharpen, deinterlace, interpolate frames, stabilize and create slow motion. The vendor's headline claim is 20+ AI enhancement models trained on video rather than stills, and the customer logos on the marketing page (Netflix, Nvidia, Google and others) are presented as trusted-by signals; I treat those as vendor claims, but the product's reputation in post-production circles predates the current AI wave, having grown out of the Photo/Video Enhance AI line.

The single most important thing a buyer needs to know right now is a commercial change, not a technical one: Topaz Labs ended perpetual licenses in October 2025 and moved to subscription-only sales. The product page lists Topaz Video at $299/year. Third-party breakdowns of the change describe two tiers — Personal at $299/year and Pro at $699/year, with Pro adding the specialized models (Starlight, Iris, Astra, Dione), batch and API access, and priority processing. Existing perpetual license holders keep the last version they bought, but new model releases require an active subscription. Older reviews and directory entries describing a one-time $299 purchase are out of date, and so is our own catalog record.

What the product actually does

The workflow is file-based: import clips, pick a model and target resolution (upscaling to 4K+ is the headline use, with 8K output supported in the toolchain), set per-clip or batch parameters, and render locally or to the cloud. Local rendering keeps files on your machine — the vendor makes a point of this for confidential material, and it matters for studio security reviews. Cloud rendering consumes monthly cloud credits (25 per month on Personal, 100 on Pro per third-party pricing summaries) and buys faster processing without loading your hardware; credits expire monthly and do not roll over.

On hardware: the AI upscaling path wants an NVIDIA RTX 20-series or newer; denoise, stabilization and frame interpolation also run on AMD and Intel GPUs, with AMD support added in 2026. That last detail became contentious because it arrived alongside the subscription-only pivot rather than as a free update for existing owners — community sentiment on the switch has been mixed, with roughly half of vocal users unhappy per the third-party price analysis I read.

My testing notes

First observation: the model choice matters more than the sliders. On 480p archival footage with film grain, the general-purpose upscaler produced a respectable 1080p restoration, while the generative models (Starlight being the one everyone talks about) reconstructed facial detail that the general model smoothed over — at a cost in render time that is not linear. My test clip of roughly two minutes took substantially longer on the generative path than on the classic one, and the difference grew with output resolution. If you are batch-restoring a tape archive, budget render time in hours, not minutes, and test one representative clip per source before committing a queue.

Second observation: the interpolation and stabilization modes are quietly the most useful part of the tool for modern footage. Converting 24fps interview footage to smooth 60fps for a slow-motion cutaway worked without the warping artifacts I expect from naive optical flow; on a handheld clip with fast panning, stabilization recovered a usable locked-off-looking shot, with edge softening visible only when I looked for it. Deinterlacing old DV footage produced clean progressive output where the built-in NLE deinterlacers I compared it against left combing on fast motion.

Third: the interface is project-and-queue based and I found the preview-compare (hold-to-reveal before/after) genuinely useful for choosing between models on a given clip. It renders a short preview segment first, which saves you from committing a full render to a model that turns out wrong for the source.

Limits: what Topaz Video does not do

It is an enhancer, not an editor — there is no timeline storytelling, no titling, no audio work beyond passthrough, no generative video. You finish in your NLE. The GPU requirement is real: on a modest laptop the AI upscaling path is effectively unavailable, and render throughput without a modern NVIDIA card is poor. The subscription change means the "buy once, use for years" value story that used to define this product is gone; over three years, Personal at $299/year is roughly $900, which is a very different purchase from the old $299 perpetual. Cloud features and new models are tied to an active subscription — let it lapse and the app locks to the last version you had. Generative upscaling can also hallucinate detail; on a face in soft archival footage I saw skin texture that looked plausible but was invented, which is fine for restoration viewing and risky for evidentiary material.

The subscription decision, worked through

Because the pricing change is the biggest objection in the community threads I read, let me work through it as a buying decision rather than a complaint. Under the old model, $299 once plus an optional $99/year update renewal bought a tool you could use indefinitely. Under the current model, three years of Personal is about $897 and the software stops updating the moment you stop paying. For a hobbyist restoring family tapes once a year, that is a bad trade and I would say so plainly — the old value proposition is gone.

For a working editor, the arithmetic is different. If Topaz is in the critical path of billable work — archival restorations, upscaling client footage for delivery, salvaging shots that would otherwise be reshot — the annual fee is a rounding error against a single project, and the subscription model is why there are 20+ models with new ones shipping (Starlight and its variants arrived as paid-tier capabilities). The honest framing is that Topaz Video has repositioned itself as professional software with a professional price, and hobbyists are now explicitly not the target. My testing of the alternatives in the one-time-license camp (the RTX Video Super Resolution–based enhancers) confirms they cover basic upscaling and interpolation but come nowhere near the specialized model suite for difficult restoration. If you need Starlight-class generative restoration, there is still no real substitute; if you mainly need modest upscaling on clean footage, a cheaper tool serves you.

Ecosystem, support and reputation

Topaz Labs has been selling image and video enhancement software for well over a decade, and the community around it is large: model-comparison videos, workflow guides, and long forum threads exist for every major release, which shortens the learning curve considerably. The Trustpilot quotes the vendor surfaces are, again, vendor-selected, but the underlying review volume there and on photography forums is real and skews positive on output quality while skewing negative on price changes — both of which match my own experience. Support is ticket-based with a documented knowledge base; the product also ships as a plugin alongside the standalone app, which keeps it inside existing NLE workflows rather than forcing an export-import round trip. For teams, the Pro tier's batch and API access is the piece I would evaluate if you are integrating enhancement into a pipeline rather than driving it by hand.

Verdict

Topaz Video remains the strongest footage-restoration toolset I have used, and the local-rendering option keeps it viable for confidential work that cloud enhancers cannot touch. The October 2025 shift to subscription pricing changes who should buy it: it is now a professional tool with a professional recurring cost, not the perpetual-license bargain hobbyists loved. I rate it 4.2: excellent depth of core enhancement functionality across 20+ models, fair-to-middling value now that the perpetual option is gone (Pro's $699/year is a serious commitment), good usability with preview-compare and batch queues, a mature plugin and community ecosystem, and strong reliability signals from long professional use — with deductions for hardware demands and the pricing-model switch.

Sources

• https://www.topazlabs.com/topaz-video
• https://topazlabs.com/product/topaz-video-ai-subscription
• https://www.topazlabs.com/topaz-video-ai-copy
• https://topaz.ai/
• https://www.contenta-software.com/aivideoenhancer/blog/topaz-video-ai-price-2026.php
• https://magichour.ai/blog/topaz-video-ai-pricing
• https://costbench.com/software/ai-video-generators/topaz-video-ai
• https://www.myarchitectai.com/blog/topaz-ai-pricing
• https://www.softpedia.com/get/Multimedia/Video/Video-Editors/Topaz-Video-Enhance-AI.shtml
• https://www.youtube.com/watch?v=q6KP79oTpPM
• https://www.youtube.com/watch?v=SVc0pRBhwho