Somebody in a photo thread will tell you Topaz Gigapixel is the only upscaler worth paying for. Two replies down, somebody else says Upscayl does the same job for free and the paid crowd just never tried it. Both people are looking at their own results, and neither is lying about what they saw. You just can’t settle Topaz Gigapixel vs Upscayl from screenshots that other people already picked to make their point.
So here’s the version without the loyalty. What each one actually is, where the free tool holds up, where the paid one still pulls ahead, and how to check which wins on the specific image in front of you instead of taking anyone’s word for it.
What you’re actually comparing
Upscayl is a free, open-source desktop app for Windows, Mac, and Linux. Under the hood it runs Real-ESRGAN plus a shelf of community models, and if you’ve read anyone rave about Real-ESRGAN as the best free upscaler, Upscayl is the friendly window bolted onto exactly that engine. It runs entirely on your own machine, no upload, no account, and it’s picked up something north of 40,000 stars on GitHub, which for a desktop tool is a lot of people bothering to click a button. There’s an optional paid cloud version if you’d rather not tax your GPU, but the app itself is free with no watermark and no cap.
Topaz Gigapixel is the paid desktop app. It uses Topaz’s own trained models, with separate ones aimed at faces, low-res source, compression cleanup, and photographic texture. The pitch has always been that those purpose-built models recover fine detail the general open-source models smear, and on the right photo you can actually see it happen.
The money side changed recently and it’s worth knowing before you get attached. Topaz dropped its old one-time license, so Gigapixel is subscription now, somewhere around $150 a year for the Personal plan depending on when you catch a sale. If you bought a perpetual license years ago you can keep running that version, you just don’t get new models. Upscayl’s price is zero and stays zero.
Where the free one is fine
For a 2x upscale headed to the web, a screen, or a normal print, Upscayl gets close enough to Topaz that paying becomes optional. The gap that used to be obvious mostly isn’t there at that size. Product shots, blog images, an old photo you want a bit sharper, AI art you’re cleaning up before posting, Upscayl handles all of it, and half the time you’d fail a blind test on which tool did which.
The part people skip is that Upscayl isn’t one thing. Switch the model and the output changes as much as switching apps would. UltraSharp leans crisp and suits illustration and AI art. Remacri sits more balanced. The general photo model is the safe default. If your first Upscayl result looked worse than Topaz, there’s a decent chance you just had the wrong model loaded for that image and never found out.
The first time I ran the same portrait through both, I’d braced for Topaz to just wipe the floor. Instead I spent five minutes dragging a slider back and forth, not sure which side I liked more, and that was the moment the “you get what you pay for” line stopped being automatic for me.
Where Topaz still earns it
Push past 2x, or hand it a face, or lean on fine text and hair, and Topaz tends to pull back ahead. Its face model in particular does a cleaner job holding a person’s actual features together instead of drifting into a smoothed stranger, and at 4x and up on a low-quality source it recovers a bit more without the texture going crunchy. If you’re doing high-res print, restoring old family photos where faces are the whole point, or upscaling for paying clients who’ll pixel-peep, that margin can be worth the subscription on its own.
The tie-breaker is boring but true: Topaz is a polished app that stays out of your way, with batch handling and a workflow built for someone doing this every day. Upscayl is good software, but it’s a smaller open-source project, and you feel that difference more in the edges of the experience than in the pixels.
What upscaling actually does
Upscaling is guessing. The detail it hands you was never in the file to begin with. Both tools invent pores, hair strands, fabric weave, and “better” mostly means the guess landed somewhere you like. That’s why the galleries lie. Every upscaler’s marketing page shows the one image where its particular hallucination happened to flatter the subject.
It also means “Topaz vs Upscayl, who wins” has no single answer. The winner flips depending on the picture. A model that’s magic on anime can turn a landscape to plastic, and the one that nails skin will sometimes wreck small text. Whatever came out best on the image in front of you is the right tool for that image. Might be the other one tomorrow.
Just test it on your own image
This part takes two minutes and beats every review, this one included. Feed both tools the exact same source file, and match the output size, a 2x and a 4x at different final dimensions isn’t a fair fight. Then drop the two results into a comparison slider and drag the handle across the same face, then a hard edge, then some patch of flat texture where a model likes to invent things.
Here’s an upscale done two ways, one softer and one sharper. Drag between them and start asking which “detail” is real and which is just the sharpen dial cranked up:
Watch a few things as you slide. Style first, sharp against smooth, which mostly comes down to taste. Then park the handle on a hard edge, a roofline against sky or dark hair against a pale wall, and look for a faint bright or dark halo hugging it. That halo is the ringing GAN-based upscalers tend to slip in. Last, stare at the smooth areas. If new detail is appearing in flat mid-tones where there was nothing to recover, the model is making it up, and you get to decide whether you mind.
Comparing three or more models at once, the multi-compare mode holds several outputs behind one link and lets you swap which pair you’re viewing, which is the sane way to rank a whole batch instead of squinting at four browser tabs.
Run that test before you commit to a download or a card number. Free might be all you need, or the paid model might save the one photo that matters, and the slider tells you which in the time it’d take to read another review.