Scaling can help you create AI-generated wall art
Generative AI art services are capable of delivering stunning images, artwork so beautiful you might want to hang it on your wall. But there’s a big problem with AI-generated images: they’re too small!
The resolution of images created by services such as ChatGPT’s DALLE or Adobe’s Firefly are often not high enough to be printed at A4, let alone anything larger.
If you want to avoid blurry artwork, you will need to level up your images. Let’s explore the different ways to improve AI art and how to turn your AI creations into poster-size prints.
Free Image Upscalers
If you type “free image upscalers” into Google, you will find dozens of online services offering to download and upgrade your images for free. However, these services often have mixed results and are subject to restrictions.
Upscale.media from Pixelbin.io is a good example. It did a passable job of scaling an AI-generated image of an oak tree that I uploaded to the service, but it blurred the details, so it’s limited to upscaling 2x scale (doubling the image dimensions) and you can only upload images smaller than 1500 x 1500 pixels in the first place.
It was a similar story with imgupscaler.com. Its free service limits you to downloading JPEG or PNG files, not WEBP files created by ChatGPT for example. There are also strict size restrictions and the results have been disappointing.
These free services can just about provide a usable result for the web, but if you want to print your AI art creations, you’ll almost certainly have to pay for professional-quality scaling.
Scaling Images in Adobe Lightroom
Adobe Lightroom can upgrade your images
It is not only a dedicated scaling software that can enlarge the size of your AI images, but also a daily photo editing software. Adobe Lightroom, for example, has a hidden feature that allows you to upgrade low-resolution images.
To access the “super resolution” feature in Adobe Lightroom Classic:
- Right-click any image in your Adobe Lightroom library
- Select Improve
- Make sure the Super Resolution box is checked in the window that appears
You don’t have too many controls to play with, but Lightroom did a decent job of doubling the dimensions of a moody seascape I created with Midjourney.
If you look at the preview image above and think it looks jarring, keep in mind that it’s at full zoom. Although Lightroom’s upscaled image contained some noisy artifacts, it did a very decent job of doubling the size of the AI art.
However, if you want to get the best results and greater control over the scaling process, you will need dedicated scaling software.
Scaling images with Topaz Gigapixel 7
Topaz Gigapixel 7 uses AI to improve detail levels in upscaled images
If you regularly improve AI art or any other type of imagery, it’s worth investing in dedicated software. Topaz Gigapixel 7 certainly isn’t cheap at $99, but it’s a professional-grade tool that delivers high-quality results.
Gigapixel 7 offers a range of AI models that you can choose from to get the best results. For example, it has a model dedicated to AI-generated art, one suitable for buildings and architecture, another for very low resolution images, and much more.
It also gives you full control over how much scaling you need. There are presets for 2x, 4x, and 6x, or you can set a specific width/height if you need to meet minimum size requirements.
In all my tests, using both real photos and AI-generated illustrations, Gigapixel 7 delivered impressive results with low levels of noise and blur (which can be controlled with sliders, by the way).
His real strength is “retaining” details. I fed it an already reasonably high resolution 2912×1632 image from Midjourney and processed a 4x upscaling. In a matter of seconds, on my MacBook Pro, it generated a huge 11,648 × 6,528 image that showed almost no signs of artificial enhancement, with details such as the individual whiskers in a man’s beard still perfectly sharp .
If you would like to view the scaled image at full resolution and see for yourself, you can download the original and scaled images here.
I used the word conservation in quotes above because, of course, there is no detail to preserve when you scale images. The AI model efficiently adds details to the image by best guessing what it should look like.
“Just purely mathematically, if you have a photo and you scale it 4x, then you have 16x more pixels,” said Eric Yang, CEO of Topaz Labs, in a briefing with me this week. “Then 15 of those 16 pixels would be generated.”
However, what we are talking about here is just an educated guess and not a complete fabrication. The Topaz Gigapixel AI “will use a lot of the surrounding context to generate it, so it won’t copy and paste elements from another image.”
Yang insists that Gigapixel AI has no more difficulty scaling AI-generated images than real photos. And there is a growing trend of people wanting to print AI-generated images at large sizes. “One of the major use cases for the Gigapixel is preparing images for printing,” he said.
“Printing seems like kind of an old-school use case. We’re actually seeing more and more people doing it because it’s getting easier to generate content. And, of course, printing requires extremely high resolution and quality, and AI upscaling is a fantastic solution for that.
And unlike real photos, where AI might invent details of someone’s face that don’t actually look like the real person, this is less of a concern with AI-generated images, where faces are faked first place.
The final advantage of using local software such as Gigapixel AI, available for Mac and Windows, is that you are not at the mercy of online services, which normally charge based on usage and often have opaque privacy policies. The AI models are downloaded to your computer, so you can upgrade as many images as you want and neither the source nor the upscaled images ever leave your system unless you decide to share them.