creative

Best AI Image Upscalers

Compare the best AI image upscalers in 2026 — Topaz Gigapixel, Magnific, Upscayl, Photoshop Generative Upscale, Let's Enhance, Pixelcut, Krea, Clipdrop, Recraft and Remini — with verified pricing, maximum output resolutions, free tiers, open-weight alternatives and per-image API costs.

Updated August 2026

Quick answer: The best AI image upscaler in 2026 for photographs is Topaz Gigapixel, because it upscales to a maximum of 32,000 pixels on the longest edge, it runs locally with no upload, and Adobe thought it valuable enough to agree to buy the company on 25 June 2026. The best upscaler for AI-generated art is Magnific, because its generative models add invented detail rather than just reconstructing what is already there. The best free option is Upscayl, which is AGPL-3.0 licensed, runs entirely offline on Windows, macOS and Linux, and applies no watermark. The one caveat that matters: every generative upscaler invents pixels that were never in the original file, so none of these tools should be used where the output has to be evidentially faithful to the source.

This guide covers tools whose primary job is increasing the pixel dimensions of an existing image. That is a different job from creating an image from a prompt, which is covered on best AI image generators, and a different job from retouching, masking and colour work, which is covered on best AI photo editors. Several products here do all three, so we say explicitly which job each one is actually good at. Every price is US list price taken from the vendor’s own page on 16 August 2026, and we say “not published” rather than repeating a third-party guess.

The current state of AI image upscaling: August 2026

Five things changed this category in the last twelve months, and the biggest of them is a consolidation.

Adobe agreed to buy Topaz Labs. Adobe announced a definitive agreement to acquire Topaz Labs on 25 June 2026, and the deal terms were not disclosed (Adobe). The transaction had not closed as of 16 August 2026; Adobe’s release states it “is expected to close in the second half of 2026, subject to the receipt of required regulatory approvals”. The US Federal Trade Commission granted early termination of the Hart-Scott-Rodino waiting period on 29 July 2026, removing the main antitrust obstacle (Yahoo Finance). Adobe has committed that “Topaz Labs products will remain available as standalone offerings through the company’s website” and that Topaz chief executive Eric Yang will continue to lead the team. Treat that commitment as a statement of intent, not a guarantee, when you are choosing what to buy today.

Perpetual licences died in this category. Topaz Labs ended perpetual licences on 3 October 2025 alongside the launch of Topaz Studio (CG Channel). Gigapixel previously sold as a one-off purchase at around $99. It now costs $149 a year on the cheapest annual plan. ON1 Resize AI 2026 is the main survivor, and it markets itself explicitly on that contrast: “We believe in perpetual licenses, no forced subscriptions.”

Freepik renamed itself after its upscaler. Freepik, which acquired Magnific in May 2024, rebranded the entire company to Magnific on 28 April 2026, folding a 250-million-asset stock library and a full generation suite under the upscaler’s name (Magnific). The old standalone Magnific plans no longer exist. The original service is still reachable at magnific.ai/legacy, which is worth knowing if you are following an older tutorial.

Upscaling became an API primitive rather than an app feature. fal.ai now resells Topaz’s entire model lineup as an endpoint, priced at $0.08 for output up to 24 megapixels. On Replicate, the philz1337x/clarity-upscaler model runs at $0.014 each, or 71 runs per dollar (Replicate). If you are upscaling in volume, the cheapest route in 2026 is almost never a subscription.

The open-weight leader changed, and the old default stopped shipping. Real-ESRGAN, still the engine inside most free upscalers, has not had a release since v0.3.0 on 20 September 2022 (GitHub). The current open-weight leader is SeedVR2 from ByteDance Seed, an ICLR 2026 diffusion transformer released under Apache 2.0 with an official ComfyUI tutorial (ComfyUI docs). That licence matters: the better-known SUPIR is published under a non-commercial-use-only declaration and cannot legally be used in a commercial product (GitHub).

One thing that did not change: Google still ships no general-purpose image upscaler. Gemini’s image models generate natively at 1K, 2K and 4K, and the Gemini image API contains no upscale operation at all (Google). Google Photos’ Zoom Enhance is limited to six Pixel models and does not work on images above 12.5 megapixels (Google).

The three types of AI image upscaler

Match the type to the job before comparing individual products. Choosing the wrong type produces an image that looks worse the closer you inspect it, and no amount of paying more fixes that.

The single most common mistake in this category is running a generative upscaler on a photograph of a person. Generative models will change the face. Use a reconstructive model with face recovery instead.

Best AI image upscalers ranked (August 2026)

Ranked by overall suitability for a general user today, weighting output quality, maximum resolution, price, free-tier usefulness and whether the tool runs locally. Prices are US list and per month unless stated.

The best AI image upscalers compared

1. Topaz Gigapixel — best overall for photographs

Type: Reconstructive, with optional generative models Price: Personal $149 paid upfront for the year, or $19 a month on a twelve-month commitment, or $29 a month rolling; Pro $499 a year; Topaz Studio bundle $399 a year; Studio Pro $799 a year (Topaz Labs) Platforms: macOS 13 and later on Apple Silicon, Windows 10 and 11, plus Photoshop and Lightroom Classic plugins. There is no Linux build. Key features: Standard, Standard Max, High Fidelity, Low Res, Text and Shapes, Art and CG, Recover, Redefine, Face Recovery and the Wonder generative models

Topaz Gigapixel is the reference implementation of photographic upscaling, and it is the tool the rest of the market is measured against. It runs locally, so files never leave your machine, and it is the only upscaler here whose output Adobe has already embedded inside Photoshop as a selectable engine.

Why it wins: Topaz Gigapixel upscales to a maximum of 32,000 pixels on the longest edge (Topaz), which at a square aspect is a little over one billion pixels — an order of magnitude beyond most web tools. Gigapixel’s model list is the broadest in the category, so a scanned document, a CGI render and a portrait each get a purpose-built model rather than one generic algorithm. Local rendering is unlimited on every plan, which makes the annual price predictable in a way credit systems are not.

Limitations: Perpetual licences ended on 3 October 2025, so a tool that cost about $99 once now costs $149 a year forever. Topaz’s own site contradicts itself on capability: the product page says “6x Lossless Zoom” while the site navigation says “up to 16x pixels”, and the page claims “nine enhancement models” while the pricing card lists eleven and the Pro page says “10+”. Do not quote a model count from Topaz marketing. Hardware demands are real: Topaz’s documentation classifies outputs above 256 megapixels as “very large upscales” requiring a minimum of 128GB of RAM (Topaz docs). Topaz does not publish free-trial terms on the Gigapixel pages; what it does publish is a two-day full refund window (Topaz docs).

Best for: Photographers producing print-size output, anyone upscaling files that must remain faithful to the original subject, and studios that need offline processing.

2. Magnific — best for AI art and generative detail

Type: Generative Price: Premium $20 a month, or $14.50 billed annually; Premium+ $45, or $33.75 annually; Pro $280, or $210 annually (Magnific) Credits: Magnific publishes annual allowances — Premium 240,000 a year (20,000 a month), Premium+ 600,000, Pro 4 million. Magnific Precision costs 90 credits per 2K image (Magnific). Key features: Magnific and Classic upscaling models, Skin Enhancer, Relight, video upscaling, and Topaz upscalers on Premium+ and above

Magnific is the tool that made generative upscaling a category. It does not try to recover what the camera saw; it re-renders the image at higher resolution and invents texture, pores, fabric weave and foliage as it goes. On an AI-generated image that is exactly what you want. On a photograph of a real person it is a liability.

Why it wins: Nothing else adds convincing invented detail as aggressively as Magnific. Since the 28 April 2026 rebrand from Freepik, one subscription now covers the upscaler alongside image, video, audio and 3D generation drawing on a shared credit balance, which makes it the best-value option if upscaling is one step in a broader creative workflow rather than the whole job. Premium+ and Pro also bundle Topaz’s upscalers inside the same subscription, so Premium+ at $33.75 a month billed annually gives you two of the top three tools in this guide on one bill.

Limitations: The Free plan has no access to the Image Upscaler at all, so there is no way to evaluate the output quality without paying. Magnific does not publish a maximum output resolution. Credits are consumed by every upscale even on the “unlimited generation” tiers, because unlimited applies to image generation and explicitly excludes Magnific upscaling. At 90 credits per 2K image, a 20,000-credit Premium month is roughly 222 upscales if you spend nothing else.

Best for: AI artists, concept designers, and anyone whose source image was generated rather than captured.

3. Upscayl — best free upscaler

Type: Reconstructive Price: Free under AGPL-3.0 for the desktop app. Upscayl Cloud Pro is $24.99 for 300 credits a month (Upscayl). Platforms: Windows 10 and later, macOS 12 and later, and Linux via Flathub, Snap, AppImage, RPM, DEB and AUR Key features: Real-ESRGAN, High Fidelity, Remacri, UltraSharp and Ultramix Balanced models, batch processing, custom model loading

Upscayl is the honest answer to “what is the best free AI image upscaler”. It is a desktop wrapper around Real-ESRGAN and several community-trained models, it processes entirely on your own GPU, it uploads nothing, and it applies no watermark. It has more than 48,000 GitHub stars (GitHub).

Why it wins: Upscayl asks for no upload, no account, no watermark and no credit limit. For a batch of a few hundred product images or screenshots, Upscayl does in an afternoon what a credit-based web tool would charge tens of dollars for. Upscayl is also the only cross-platform desktop option in this guide, and the only one with a Linux build.

Limitations: Upscayl requires a Vulkan-compatible GPU, so most integrated graphics and all CPU-only machines cannot run it. Development has effectively stalled: the latest release is v2.15.0 from 25 December 2024, roughly twenty months old as of 16 August 2026, and Upscayl still ships Real-ESRGAN models that were last updated in 2022. Upscayl’s own site claims enhancement “by up to 16x better resolution”, which describes pixel count rather than a linear scale factor, and neither the site nor the repository publishes a maximum output resolution. The “completely free” framing is also no longer strictly true — there is a paid Upscayl Cloud tier and the Mac App Store build is a paid download, though the version from the website remains free.

Best for: Anyone with a discrete GPU who wants unlimited, private, zero-cost upscaling and does not need generative detail.

4. Adobe Photoshop Generative Upscale — best if you already pay Adobe

Type: Both, depending on model selected Price: Photoshop single app $22.99 a month billed annually with 25 generative credits; Photography plan $19.99 with 1,000 credits; Firefly Pro $19.99 with 4,000 credits; Creative Cloud Pro $69.99 with 4,000 credits (Adobe) Scale factors: 2x or 4x only Key features: Three selectable engines — Firefly Upscaler, Topaz Gigapixel and Topaz Bloom — plus a face recovery toggle when Gigapixel is selected

Adobe shipped Generative Upscale in Photoshop 2026 and, unusually, did not restrict it to its own model. Adobe’s documentation states: “From the Scale menu, select 2x or 4x to define how much you want the image to be scaled” (Adobe). The same panel lets you pick Topaz Gigapixel to preserve existing detail or Topaz Bloom to add new creative detail, which is the clearest possible signal about where the acquisition is heading.

Why it wins: Photoshop Generative Upscale is the only upscaler in this guide that sits inside the editing workflow rather than beside it, and the only one offering both a reconstructive and a generative Topaz model without a second subscription. If you already pay for Photoshop or the Photography plan, your marginal cost of upscaling is zero until the generative credits run out.

Limitations: 2x and 4x are the only options, which is far short of Topaz’s 32,000-pixel ceiling or Let’s Enhance’s 512 megapixels. The Firefly Upscaler tops out at 6144 x 6144 pixels. Adobe does not publish a per-image generative credit cost, stating only that “credit consumption varies by model”, so the effective price per upscale is unknowable in advance. The Photoshop single-app plan includes just 25 generative credits a month, which is close to unusable for upscaling at volume; the $19.99 Photography plan’s 1,000 credits is the better buy for this specific job.

Best for: Photoshop and Lightroom users upscaling occasionally as part of a larger edit.

5. Let’s Enhance — best for the largest output files

Type: Reconstructive Price: Starter $9 a month billed annually, or $12 rolling, for 100 credits; Pro $24 or $32 for 300 credits; Max $34 or $45 for 500 credits (Let’s Enhance) Maximum output: 256 megapixels on Starter, 350 on Pro, 512 on Max Key features: One image equals one credit, credit rollover up to six times the monthly allowance, batch processing, background removal

Let’s Enhance has the highest published output ceiling of any consumer web upscaler in this guide. 512 megapixels is roughly thirty times Remini’s 4096 x 4096 limit, and the two products are marketed with the same words.

Why it wins: A flat one-image-one-credit model is the easiest pricing in the category to reason about, and unused credits roll over while you remain subscribed. For billboard, exhibition or large-format print work, the 350 and 512 megapixel tiers have almost no competition outside desktop software.

Limitations: The free tier gives 10 one-off credits, caps output at 8 megapixels and watermarks the result, though re-downloading after upgrading removes the watermark retroactively. Input is capped at 64 megapixels or 50MB on paid plans. Let’s Enhance also contradicts itself: the plan cards state 256, 350 and 512 megapixels while the FAQ on the same site says 256 for personal plans and 500 for business. The plan cards are the figures the checkout actually sells.

Best for: Large-format print, exhibition work, and anyone who needs a single very large file rather than many small ones.

6. Pixelcut — best free web upscaler

Type: Both, via Fast Upscale and Creative Upscale modes Price: Pro $10 a month, or $8 billed yearly, for 600 credits and unlimited upscaling; Business $30 or $24 for 3,600 credits (Pixelcut) Maximum output: 16K Key features: Free export with no watermark and no signup, bulk upscaling of up to 10,000 images on Pro, Creativity and Resemblance sliders

Pixelcut is the only tool in this guide with a genuinely unrestricted free web tier: no watermark, no account, and no low-resolution download trick. That alone makes it the default recommendation for someone with one image and no budget.

Why it wins: Pixelcut’s free output is clean and full-resolution. On Pixelcut’s paid plans upscaling itself is unlimited and credits are only consumed by third-party frontier models, which inverts the usual credit anxiety. Bulk export of up to 10,000 images at $8 a month works out at $0.0008 an image, which undercuts every API in this guide for anyone willing to run batches through a browser.

Limitations: Input is capped at 100MB and 8000 x 8000 pixels, so it cannot take a large file and make it larger the way Topaz can. It is built around e-commerce and product photography, so the tooling around the upscaler assumes that use case. Creative Upscale is generative and will alter faces.

Best for: One-off free upscales, e-commerce catalogues, and high-volume batch work on a small budget.

7. Krea — best inside a generation workflow

Type: Generative Price: Basic $9 a month, or $5.25 billed annually, for 5,000 compute units; Pro $35 or $21 for 20,000; Max $105 or $63 for 60,000; Business $200 or $160 (Krea) Maximum output: Up to 22K on paid plans; the free Enhancer tool is documented at 8K Key features: Enhancer, real-time generation, video upscaling, and the in-house Krea 2 image model

Krea is a generation platform whose Enhancer happens to be excellent, rather than an upscaler with generation bolted on. If you are already producing images in Krea, upscaling in the same canvas with the same compute balance removes a step.

Why it wins: Krea’s 22K ceiling on paid plans is the highest published figure of any web tool in this guide, and the Krea Enhancer inherits Krea’s aesthetic tuning, which produces more coherent invented detail than a generic diffusion upscaler. The $5.25 annual Basic tier is the cheapest serious paid entry point in this guide.

Limitations: Krea publishes two different ceilings — 22K in the paid plan feature list and 8K on the Enhancer landing page — without explaining the difference, so treat 8K as the free tool’s limit and 22K as the in-app paid limit. The free plan’s 100 compute units per day is roughly one image generation, so it is a demonstration rather than a usable tier. Compute unit costs vary by model and resolution, which makes per-image cost hard to predict.

Best for: Designers and artists already generating in Krea, and anyone who wants very high resolution generative output at a low monthly price.

8. Recraft — best cheap upscaling API

Type: Both, via Crisp Upscale and Creative Upscale Price: Basic $12 a month, or $10 billed annually, for 1,000 credits; Pro tiers from $20 a month for 2,000 credits up to $160, or $128 billed annually, for 16,000 (Recraft) Credit cost: Crisp Upscale 1 credit; Creative Upscale 20 credits (Recraft) API price: Crisp Upscale $0.004 per image; Creative Upscale $0.25 per image (Recraft)

Recraft has the cleanest separation between the two jobs in this category, and prices them accordingly. Crisp Upscale costs twenty times less than Creative Upscale in credits and sixty times less through the API, which is an unusually honest reflection of how much compute each actually needs.

Why it wins: At $0.004 per crisp upscale via the API, Recraft has the cheapest published per-image API price of any commercial provider in this guide. For a pipeline that needs clean 4x output on thousands of assets, nothing else on a subscription model comes close.

Limitations: Recraft does not publish a maximum output resolution for either upscaler anywhere in its documentation; its own blog implies 4x but that is marketing copy, not a specification. On the Free plan Recraft owns the images you generate, they appear publicly in the community gallery, and they carry no commercial licence — paid plans grant full ownership, retained after cancellation. Subscription credits do not roll over.

Best for: Developers building upscaling into a product, and design teams working at volume.

9. Clipdrop — best free x2 in a browser

Type: Generative Price: Free tier gives x2 upscaling with 20 uses per 24 hours; Pro gives x16 with 1,000 uses per 24 hours. Clipdrop does not render a price on its pricing page, so we do not publish one (Clipdrop). Owner: Jasper. Clipdrop was built by InitML, acquired by Stability AI in 2023 and sold to Jasper in 2024; the site now states “Clipdrop is part of Jasper now”. Key features: x2, x4, x8 and x16 options in the interface, plus cleanup, relight, background removal and text removal

Clipdrop survives as a fast, no-friction browser tool. Note that its own marketing copy and FAQ are stale and still claim “up to 4 times”, while the live scale selector offers x2, x4, x8 and x16. Trust the interface, not the FAQ.

Why it wins: Clipdrop gives twenty free x2 upscales a day with no install and no watermark, which is a genuinely useful free allowance, and the surrounding toolset means one tab handles upscale, cleanup and background removal.

Limitations: The pricing page serves placeholder text rather than a figure, which is a poor signal about how actively the product is maintained. Third-party sources quote Pro prices between $9 and $16 and disagree with each other, so we have not published a number. Upscaling here is generative and will alter faces.

Best for: Quick free upscales in a browser, and anyone already inside the Jasper ecosystem.

10. ON1 Resize AI 2026 — best perpetual licence

Type: Reconstructive Price: One-time perpetual purchase covering two computers, with a 30-day money-back guarantee. ON1 renders its prices in JavaScript and served no figure, so we do not publish one (ON1). Scale factor: “700% or more” per ON1’s own store copy Key features: Highest Quality diffusion model and Standard fast model, improved face recovery, plugin for Photoshop, Lightroom Classic and Capture One

ON1 Resize AI 2026 exists for one reason and states it plainly: “21 Years, Photographer First — We believe in perpetual licenses, no forced subscriptions.” In a category where the market leader abandoned perpetual licensing in October 2025, that is a real differentiator rather than a slogan.

Why it wins: You buy ON1 Resize AI 2026 once, run it on two machines, and receive future updates to that version. Layered PSD support and a Capture One plugin make it the best fit for photographers whose workflow does not run through Adobe.

Limitations: ON1’s own store page reports “700% or more” while third-party reviews claim up to 10x — we have cited ON1’s figure. Windows on ARM is not supported. ON1 ships two upscaling models against Gigapixel’s much broader list, so there is no specialised handling for text, CGI or line art.

Best for: Photographers who refuse subscriptions, and Capture One users.

11. Remini — best for old and blurry photos of people

Type: Restorative Price: Weekly and monthly subscriptions sold through the app stores. Remini publishes no price list and runs multiple concurrent price tests, so any single figure would be wrong for most readers (App Store). Maximum output: 2x, capped at 4096 x 4096 pixels (Remini) Owner: AI Creativity S.r.l., a Bending Spoons company

Remini is not really an upscaler and should not be judged as one. It is a face and photo restoration tool that happens to enlarge, and it is the best consumer option for a damaged family photograph.

Why it wins: Remini reconstructs faces from low-quality source material better than anything else at consumer price, and Remini’s mobile-first flow is the easiest in this guide for a non-technical user.

Limitations: The 2x, 4096 x 4096 pixel ceiling is roughly thirty times smaller than Let’s Enhance’s 512-megapixel maximum, which is the single clearest illustration of how loosely the phrase “AI upscaler” is used. The free tier is ad-gated and watermarks output unless you watch further ads. Pricing is weekly-first, region-dependent and actively A/B tested — the UK App Store listing alone shows four different weekly prices for the same tier.

Best for: Restoring old, scanned or blurry photographs of people on a phone.

Open-weight and self-hosted upscalers

If you can run a GPU, the open-weight options are free, unlimited and in one case better than the commercial tools. The licence is the thing to check first, because two of the most famous models cannot legally be used commercially.

The practical way most people run these is ComfyUI, which raised $30 million at a $500 million valuation on 24 April 2026 and reports more than four million users and 60,000 community nodes (TechCrunch).

The short version: use SeedVR2 if you need commercial rights and quality, Real-ESRGAN if you need speed and simplicity, and avoid SUPIR entirely in any commercial context regardless of how good the sample images look.

Upscaling APIs: what it actually costs per image

If you are upscaling programmatically, subscriptions are the wrong pricing model. These are published per-image or per-megapixel prices as of 16 August 2026.

The pricing insight that matters: a 24-megapixel Gigapixel upscale costs about $0.08 through an API and about $149 a year through a subscription. If you run fewer than roughly 1,800 images a year, the API is cheaper. Above that, the desktop licence wins — and it also means your files never leave your machine.

Use-case recommendations

For upscaling photographs for print

Winner: Topaz Gigapixel, $149 a year

Topaz Gigapixel upscales to 32,000 pixels on the longest edge, processes locally, and ships a High Fidelity model built for exactly this. Alternative: Let’s Enhance Max at $34 a month billed annually if you need 512 megapixels and do not want to buy software.

For upscaling AI-generated art

Winner: Magnific Premium+, $33.75 a month billed annually

Generative detail is the whole point on a generated source, and Premium+ bundles the Topaz upscalers as well. Alternative: Krea Basic at $5.25 a month billed annually if you are already generating there.

For upscaling with zero budget

Winner: Upscayl, free

Upscayl applies no watermark, uploads nothing and imposes no limit, provided you have a Vulkan-capable GPU. Alternative: Pixelcut if you do not, because its free web tier has no watermark and no signup either.

For restoring an old family photograph

Winner: Remini

Face reconstruction on degraded sources is Remini’s entire specialism, and Remini’s 2x ceiling is rarely a problem on a scan. Alternative: Topaz Gigapixel’s Recover and Face Recovery models if you already own it.

For a single image, right now, in a browser

Winner: Pixelcut, free

Pixelcut gives a full-resolution download with no watermark and no account. Alternative: Clipdrop free tier, which gives 20 x2 upscales per 24 hours.

For upscaling thousands of images

Winner: Recraft Crisp Upscale API at $0.004 per image

Ten thousand images through Recraft’s Crisp Upscale API costs $40, and Pixelcut Pro is cheaper still at $8 a month for up to 10,000 images if you can run the batches in a browser. Alternative: Upscayl batch mode on your own hardware for $0, if the images are not on a deadline.

For anyone who will not pay a subscription

Winner: ON1 Resize AI 2026

The main perpetual-licence photo upscaler still being actively sold, covering two computers with a 30-day money-back guarantee. Alternative: Upscayl, which is free forever under AGPL-3.0.

For upscaling inside an existing edit

Winner: Adobe Photoshop Generative Upscale on the $19.99 Photography plan

The Photography plan gives three upscaling engines including two Topaz models, no context switch, and 1,000 generative credits a month. Note: the $22.99 Photoshop single-app plan includes only 25 generative credits a month, which is the wrong plan for this job.

For building upscaling into a product

Winner: fal.ai’s Topaz endpoint at $0.08 up to 24MP

The fal.ai Topaz endpoint gives commercial licensing and Topaz quality with no infrastructure to run. Alternative: self-host SeedVR2 under Apache 2.0 if your volume justifies the GPU.

For output that must be faithful to the source

Winner: none of the generative tools.

Use a reconstructive model — Topaz Standard or High Fidelity, Real-ESRGAN, or Adobe’s Firefly Upscaler — and document which one you used. See the section below on why this matters.

What AI upscaling actually does, and where it fails

Every tool in this guide is inventing pixels. A 4x upscale of a 1000 x 1000 image produces 16 million pixels from one million, which means fifteen out of every sixteen pixels in the output were never captured. Reconstructive models infer those pixels from local structure and stay close to plausible. Generative models synthesise them from a learned prior, which is why they look better and are less trustworthy.

Three practical consequences follow.

Text and faces are where it shows. Generative upscalers rewrite small text into convincing gibberish and shift facial features enough to change who someone looks like. This is why Topaz ships a separate Text and Shapes model and a Face Recovery toggle, and why Adobe exposes face recovery as an explicit option when the Gigapixel engine is selected.

Upscaled output is not evidence. An upscaled image cannot establish what a number plate said, what a document read, or who was in a photograph, because the detail that appears to answer the question was generated by the model rather than recovered from the file. No tool in this category should be used to support a factual claim about the contents of the original image.

Generating large often beats upscaling. Gemini’s image models now generate natively at 1K, 2K and 4K, controlled by an image_size parameter, and Google publishes no upscale endpoint at all (Google). If the image does not exist yet, generating it at the resolution you need produces a better result than generating small and enlarging.

Pricing comparison: what you’ll actually pay

Annual pricing where a discount is offered, US list prices as of 16 August 2026.

Cost strategy: if you upscale occasionally, use Pixelcut or Upscayl and pay nothing. If you upscale photographs professionally, buy Topaz Gigapixel Personal at $149 a year, because unlimited local rendering removes per-image cost entirely. If you upscale AI art, Magnific Premium+ at $33.75 a month billed annually is the only plan here that bundles two of the top three tools. If you upscale programmatically, ignore every subscription on this page and pay $0.004 to $0.08 an image for reconstructive upscaling through an API, or up to $0.25 for a generative one.

Related guides

If you need to create the image rather than enlarge it, see best AI image generators. If the job is retouching, masking, background work or colour, see best AI photo editors. For professional portraits generated from scratch, see best AI headshot generators. For the models behind these tools, see best AI models, and for how we assess and rank tools, see how we rank.

Frequently asked questions

What is the best AI image upscaler in 2026?

Topaz Gigapixel is the best AI image upscaler in 2026 for photographs, because it upscales to a maximum of 32,000 pixels on the longest edge, processes locally with no upload, and offers purpose-built models for portraits, text, CGI and low-resolution sources. Topaz Gigapixel costs $149 paid upfront for a year on the cheapest plan. For AI-generated art, Magnific is the better choice because its models add invented detail rather than reconstructing existing detail, and for free upscaling Upscayl is the strongest option because it runs offline on your own GPU with no watermark.

What is the best free AI image upscaler?

Upscayl is the best free AI image upscaler, because it is licensed under AGPL-3.0, runs entirely on your own machine on Windows, macOS and Linux, applies no watermark, and has no usage limit. Upscayl requires a Vulkan-compatible GPU, so it will not run on most integrated graphics or CPU-only machines. If your hardware cannot run it, Pixelcut is the best free browser alternative, because its free tier exports full-resolution images with no watermark and no signup.

Can AI upscaling add real detail to a photo?

No — AI upscaling cannot add real detail, because the detail was never captured in the file. A 4x upscale produces sixteen pixels for every one in the original, meaning fifteen out of every sixteen pixels in the output are generated rather than recovered. Reconstructive models such as Topaz Standard and Real-ESRGAN infer plausible detail from surrounding structure, and generative models such as Magnific and Topaz Redefine synthesise entirely new detail from a learned prior. Both produce invented pixels, which is why upscaled images cannot be used to establish what was actually in a scene.

How much does Topaz Gigapixel cost in 2026?

Topaz Gigapixel costs $149 paid upfront for a year on the Personal plan, or $19 a month on a twelve-month commitment billed monthly, or $29 a month on a rolling plan, according to Topaz Labs’ own pricing as of 16 August 2026. The Pro tier costs $499 a year and adds full commercial use, seat management and local access to the Wonder 1 model. Perpetual licences are no longer available: Topaz ended them on 3 October 2025, having previously sold Gigapixel outright for about $99.

Is Adobe buying Topaz Labs?

Adobe announced a definitive agreement to acquire Topaz Labs on 25 June 2026, and the deal had not closed as of 16 August 2026. The financial terms were not disclosed. Adobe stated that the transaction is expected to close in the second half of 2026 subject to regulatory approval, and the US Federal Trade Commission granted early termination of the antitrust waiting period on 29 July 2026. Adobe has said Topaz products will remain available as standalone purchases through Topaz’s own website after close, and that Topaz chief executive Eric Yang will continue to lead the team.

What is the maximum resolution an AI upscaler can produce?

Let’s Enhance publishes the highest ceiling among consumer web upscalers at 512 megapixels on its Max plan, while Topaz Gigapixel publishes the highest ceiling among desktop tools at 32,000 pixels on the longest edge, a little over one billion pixels at a square aspect. Krea documents 22K output on paid plans and Pixelcut documents 16K. At the other end of the same category, Remini caps output at 4096 by 4096 pixels, about 16.8 megapixels, which is roughly thirty times smaller than Let’s Enhance’s maximum despite both products being marketed as AI upscalers.

Should I upscale an image or generate it at a higher resolution?

Generate it at a higher resolution if the image does not exist yet, because native generation produces genuine detail while upscaling produces inferred detail. Google’s Gemini image models generate natively at 1K, 2K and 4K via an image_size parameter, and Google publishes no upscale endpoint at all. Upscaling is the right choice only when you are working from an image that already exists — a photograph, a scan, an old render, or a generation from a model that could not output the size you need.

Does AI upscaling work on faces?

AI upscaling works on faces only if you use a model designed for them, because generative upscalers reliably alter facial features enough to change who a person appears to be. Topaz Gigapixel ships a dedicated Face Recovery model, Adobe exposes a face recovery toggle when the Topaz Gigapixel engine is selected in Photoshop’s Generative Upscale, and Remini is built almost entirely around face reconstruction. Running Magnific, Krea Enhancer, Clipdrop or Recraft Creative Upscale on a photograph of a real person will produce a sharper image of a subtly different face.

Are open-source AI upscalers good enough to replace paid tools?

Open-weight upscalers are good enough to replace paid tools for most jobs, but the licence matters more than the quality. SeedVR2 from ByteDance Seed is released under Apache 2.0 and can be used commercially, and it is the current open-weight quality leader with an official ComfyUI tutorial. Real-ESRGAN is BSD-3-Clause and free for commercial use, but has not had a release since September 2022. SUPIR, despite being the best-known open restoration model, is published under a non-commercial-use-only declaration and cannot legally be used in a commercial product without written permission from its authors.

What is the cheapest way to upscale images in bulk?

The cheapest commercial route is Recraft’s Crisp Upscale API at $0.004 per image, which puts ten thousand upscales at $40. Replicate’s clarity-upscaler costs $0.014 per run and fal.ai’s Topaz endpoint costs $0.08 for output up to 24 megapixels. The cheapest route overall is running Upscayl or SeedVR2 on your own GPU, which costs nothing beyond electricity but requires hardware and time. A Topaz Gigapixel Personal subscription at $149 a year beats API pricing once you pass roughly 1,800 images a year.