The best AI upscaler depends on your files and where you want the work to run. Topaz Gigapixel and Topaz Video are the main paid desktop options if you want local processing and up to 16x image enlargement. Magnific is a cloud "creative" upscaler that adds new detail on purpose. Upscayl (built on Real-ESRGAN) and Video2X are free and open source if you have a suitable GPU. Kenerate AI (our product) runs seven image upscalers and a video upscaler in the browser, priced per image or per second. Whatever you pick, remember that AI upscaling invents detail. It doesn't recover detail that was never captured.

Disclosure: Kenerate AI is our product. This guide has no scores, rankings or "winner" badges. We compare tools on criteria you can check: maximum scale or output size, price, local vs cloud, and what each tool is designed for. Vendor facts come from each vendor's own pricing pages, docs or GitHub repositories, read on 23 October 2026. Kenerate figures come from the app's upscaler config on the same date.

What should you look for in an AI upscaler?

  1. Max scale or output size. Some tools quote a factor (4x, 16x), others a target (4K, 8K, 128 MP). For print, work back from the size you need.
  2. Faithful vs creative. Faithful upscalers sharpen what's there. Creative upscalers generate new texture, which can look great on AI art and wrong on a real face.
  3. Local vs cloud. Local tools keep files on your machine and have no per-image cost, but need a capable GPU. Cloud tools work on any device and charge per job or by subscription.
  4. Price model. One-off credits, a monthly plan or free.
  5. Image, video or both. Video upscaling is a separate problem: detail has to stay consistent from frame to frame.

How do the best AI image upscalers compare?

ToolRunsMax scale / outputPrice (read 23 Oct 2026)Style
Topaz GigapixelLocal (Mac, Windows) + cloud"Up to 16x"$29/mo plan, or $19/mo single app on an annual commitment ($149/yr); Pro licence $499/yrFaithful, with generative models
Topaz for Web – ImageCloud32 MP export (256 MP on Web Pro)$12/mo promo ($19 list)Faithful + generative
MagnificCloud2x, 4x, 8x or 16x; API output capped at 25.3 MPCredit-based subscription (see its pricing page)Creative, with creativity / HDR / resemblance sliders
UpscaylLocal (Linux, macOS 12+, Windows 10+)Model-dependentFree, open source (AGPL-3.0)Faithful (Real-ESRGAN and community models)
Real-ESRGAN (raw)Local (Python or NCNN builds)4x models; any output scale via the Python scriptFree, open source (BSD-3-Clause)Faithful
Kenerate AI (our product)Cloud, in the browserUp to 8K or 128 MP depending on engine0.72 to 75 credits per imageFaithful, creative and restore engines

Topaz Gigapixel

Topaz Labs' Gigapixel page advertises image upscaling "up to 16x" and runs on Mac and Windows, with both local and cloud rendering. The pricing page lists Gigapixel at $29 a month, or $19 a month as a single app on an annual commitment ($149 a year). Standard licences come with "limited commercial use". Full commercial use needs a Pro licence: Gigapixel Pro is $54 a month on an annual commitment, or $499 a year. Topaz now sells subscriptions, so budget for a recurring cost. If you'd rather not install anything, Topaz for Web – Image runs in the browser with a 32 MP export cap.

Good fit: photographers who upscale often, want files to stay local, and have a recent GPU or Apple Silicon Mac.

Magnific

Magnific is the best-known creative upscaler. Its API documentation lists scale factors of 2x, 4x, 8x and 16x. It has three sliders from −10 to 10: creativity ("increase or decrease AI's creativity"), HDR (definition and detail) and resemblance (how close the result stays to the original). API output is capped at 25.3 million pixels. Magnific is sold as a credit-based subscription that also covers image and video generation tools. On 23 October 2026 its pricing page showed us local-currency prices only, so check the current plan in your own currency before you subscribe.

Good fit: AI art, concept art and renders where you want new texture. Be careful on real faces and products: a high creativity setting will change them.

Upscayl and Real-ESRGAN (free, open source)

Upscayl is a free desktop app for Linux, macOS and Windows, licensed AGPL-3.0. It uses Real-ESRGAN and community models such as UltraSharp and Remacri. It needs a Vulkan-compatible GPU, and the project notes that many integrated GPUs won't work. Real-ESRGAN is the open-source model underneath, licensed BSD-3-Clause. Its default output is 4x, the Python script can scale to any size, and an optional GFPGAN face pass is available.

Good fit: anyone with a decent graphics card who wants unlimited, private, free upscaling and doesn't mind a bit of setup.

Kenerate AI image upscaler (our product)

Kenerate's AI image upscaler puts seven engines in one tool. Credits come straight from the app's upscaler config:

EngineTypeOutputCredits per image
Real-ESRGAN (default)FaithfulFixed 2048 px long side0.72
Kenerate UpscaleFaithful4 to 128 MP target1.5 (4 MP) to 36 (128 MP)
Image UpscalerFaithful2K, 4K or 8K3
Ultimate UpscalerFaithful2K, 4K or 8K18
Clarity ProCreative, with a creativity dial2 to 64 MP9 per output megapixel (36 at 4 MP)
Phota EnhanceRestore old or damaged photos2048 px39
Recraft CreativeCreative texture4096 px75

New accounts get 25 free credits on sign-up, enough to try every engine on your own picture, and outputs have no watermark. Credit packs start at $15 for 1,300 credits. One limit to know: the default Real-ESRGAN engine always returns 2048 px, so pictures larger than that come back smaller. Use Image Upscaler or Kenerate Upscale to go to 4K and beyond. Our guide to upscaling an image to 4K walks through it.

How do the best AI video upscalers compare?

ToolRunsMax outputPrice (read 23 Oct 2026)Notes
Topaz VideoLocal + cloud creditsModels "up to 4K" ("4K+" use case)$59/mo, or $39/mo on an annual commitment; $299/yr single appIncludes 25 cloud video credits a month; limited commercial use on the standard licence
Video2XLocal (Windows, Linux; macOS via container)Model-dependentFree, open source (AGPL-3.0)Real-ESRGAN, Real-CUGAN, Anime4K and RIFE frame interpolation; Vulkan GPU and AVX2 CPU required
Kenerate AI Video Upscaler (our product)Cloud720p, 1080p, 2K or 4K6 / 9 / 15 / 24 credits per secondClips up to 5 minutes and 100 MB

Topaz Video is the long-standing desktop choice. It renders locally without limits ("your files never leave your computer"), and cloud renders use credits. Video2X is the free route. It wraps several open models, including RIFE for frame interpolation, and can run on Google Colab if you don't have a GPU.

On Kenerate, the AI video upscaler bills by the second, rounded up. A 10-second clip costs 90 credits to 1080p or 240 credits to 4K. It takes clips up to 5 minutes, so a full 5-minute clip to 4K costs 7,200 credits. That makes it a good fit for short social clips and AI-generated video. For long footage you'll redo often, a local tool will cost less.

Does AI upscaling add real detail?

No, and it's worth being honest about this. An upscaler can't recover information the camera never recorded. It predicts plausible pixels from patterns learned in training. Faithful models like Real-ESRGAN make conservative guesses: cleaner edges, less noise, sharper texture. Creative models make bolder guesses, which is why Magnific gives you a creativity slider and a resemblance slider.

That matters most on faces. In our own tests of Clarity Pro on portraits, only a strongly negative creativity setting (−6 at 4 MP) kept the same person. Settings of 0 and above, or 8 MP, changed age, hair and even glasses. Kenerate now starts Clarity at −6 for that reason. Use these rules of thumb:

  • Faces, documents, evidence, product photos: faithful engines only, low or no creativity.
  • AI art, game assets, concept art: creative engines are fine. Compare the result with the original at 100% zoom.
  • Old or damaged photos: a restore model rebuilds missing detail, so treat the result as a reconstruction, not a scan.
  • Video: check motion. Detail that shimmers or "boils" between frames is a sign the model is guessing differently on each frame.

Which AI upscaler should you choose?

  1. You want it free and private, and you have a GPU: Upscayl for images, Video2X for video.
  2. You upscale photos professionally and want local files: Topaz Gigapixel (and Topaz Video for footage). Get a Pro licence if you need full commercial use.
  3. You want AI art to gain new texture: Magnific, or Clarity Pro and Recraft Creative on Kenerate.
  4. You want no install, pay per job, and image and video in one place: Kenerate AI. Start with the default engine, then move to Image Upscaler (3 credits, up to 8K) for bigger output.
  5. You're upscaling anime: use a model trained for it. Real-ESRGAN has an anime video model, and our anime upscaler page covers the workflow.

For more alternatives side by side, see our image upscaler alternatives page. If your video is blurry or dark rather than just small, read our guide to enhancing and sharpening AI video.

Bottom line

There is no single best AI upscaler, only the right one for the file. Free local tools (Upscayl, Real-ESRGAN, Video2X) cost nothing but need a GPU. Topaz is the paid desktop choice with up to 16x image scaling. Magnific is for deliberately creative upscales. Kenerate AI covers images (0.72 to 75 credits, up to 8K or 128 MP) and video (6 to 24 credits per second, up to 4K) in the browser. Keep faces and real products on faithful settings, because every AI upscaler is guessing at the detail it adds.

Updated 23 October 2026. Vendor prices and limits were read from each vendor's own pages on that date. Kenerate figures are from the app's upscaler config.