Today, we return to the topic that's been on every AI artist’s mind and screen since the beginning of the text-to-image neuro-revolution—turning an AI generated digital art piece into a physical high resolution print.
In my artistic career as a photographer, upscaling was a rare necessity: thanks to modern digital sensors, printing was straightforward, almost out-of-the-box. But for our Midjourney masterpieces—for postcards, posters, books, or large-format high-quality fine-art prints—isn't just an option, it's almost a requirement!
The latest model of Midjourney—V6—has two built-in upscalers: Subtle and Creative, which we covered in-depth in a dedicated guide↗︎.
However, they both have their limitations, and here are the main ones:
So in this guide, we will be reviewing popular AI upscalers that eliminate all (or most) of these limitations. Of course, there are too many such tools to cover in one go, so let’s start with spotlighting five standout solutions; including online services and downloadable apps—each tested on four different samples of Midjourney generations.
First, let’s meet our test subjects—all artworks created with Midjourney AI V6, and initially upscaled with Midjourney’s Subtle or Creative upscalers to maximize the original’s quality and size.
Where possible, we used the standard or the most general upscaling options and the model’s default or recommended settings.
All test are done on MacBook Air 13.6" Apple M2 (8C CPU/GPU), 8 Gb.
The reigning champion from our last study, and Midlibrary's top pick, is none other than Gigapixel AI. This powerhouse of an upscaler operates on a one-time purchase license and keeps itself fresh with regular model updates.
Gigapixel AI has a sleek, minimalist interface, where every control is intuitively placed, making learning quick and easy, and the user experience smooth.
Gigapixel’s Auto mode intelligently selects the best model and settings for your images, but manual control is just a click away, including options like Face Recovery, De-noise and De-blur, Fix Compression, and Gamma Correction to fix color/gamma issues.
Gigapixel AI features eight different upscaling models for a range of general and specific applications: from magnifying extremely low-resolution, compressed images to upscaling line-art, text, AI-generated art, and even ultra HD photographs, multiplying the size of already large pictures.
Notably, Standard and High Fidelity modes proved to be very versatile and in 9 out of 10 cases enough for most AI-generated pictures: from landscapes to line-art.
Super-large upscales across all models, highest quality, lighting fast.
Learn more at Gigapixel AI official website↗︎
ON1 Resize AI is a powerful tool designed for image upscaling and enchancement, with strong focus on final media.
ON1 Resize AI’s interface is elegant and—with a very small learning curve—intuitive and straightforward. Although at the beginning, the number of options might occasionally stun an inexperienced user.
There are four models to choose from for your upscale: Resize AI 2023 (Smooth), Resize AI 2023 (Faithful), Genuine Fractals, and Bicubic, each with a specific set of controls (or absence of any in the case of Bicubic).
I’ll quote the developers here:
There are plenty of options for you to choose from when it comes to the desired output size: from longest side size to magnification in percents. The recommended multiplier is 600%, however you can set this counter much, much higher. It reacts to the size of the initial image, and for our samples allowed to go as high as to 2853%, with an output picture of a gargantuan 83049x36544 size. I didn’t want to risk saving such an enormous file as PNG, but even with a 100% quality JPEG was estimated to be 28.8 Gb.
One of ON1’s standout features is its set of Printing tools:
As a cherry on top, ON1 Resize AI features batch processing, advanced export dialog, and highly customizable Export Presets allowing to have multiple version of upscales with different settings in the same export. And it only comes natural that you can export to JPG and PNG, and more professional image fomats: like TIFF, DNG, and even Adobe Photoshop PSD/PSB.
Extremely large upscales, very high quality, lighting fast + additional tools specifically for printing. Learn more at ON1 official website↗︎
LetsEnhance.io is the most versatile online-only app on our list, suited for reviving old photos, improving low-resolution pictures for marketing or product photography, or enlarging AI-generated images for printing. Its features include fixing blurry or pixelated images, noise removal, and color correction, making images vibrant and lifelike.
LetsEnhance features a number of models designed for different applications: like Photo (with Enhance Face Details option), Digital art (for illustrations, paintings, anime, etc.), Smart Enhance (for low resolution images for the web), Smart resize (recommended for images with text).
The fifth model is now in beta stage and is called Creative — a trendy (see Magnific.AI, Leonardo AI, and Midjourney’s Upscale (Creative)) mode that not only upscales an image, but also adds new details to the enlarged picture.
LetsEnhance.io also offers three presets to quickly activate a specific set of parameters: Printing (that optimizes images for printing in highest quality), E-commerce, and Real Estate.
Looking at presets, one can assume that Letsenhance specifically targets ecommerce applications, rather than fine-art printing projects.
High quality, easy to use, and fairly advanced online upscaler. Learn more at LetsEnhance.io official website↗︎
Nero AI is a web-based upscaler (and other tools, like Image Compressor, Photo Restoration, Photo Colorizer, and Image Denoiser). In the scope of this guide, we will be looking only at Nero AI Image Upscaler.
Nero AI Image Upscaler features a range of upscaler models designed for various subjects/image types (I will quote the deveolper here):
Despite its easy-to-use interface, the overall experience might seem uncomfortable due to all the unrelated elements on screen: from seasonal promotions to marketing stats and newsletter promo—making the interface cluttered and distracting.
Fair quality, medium speed, easy to use online upscaler. Learn more at Nero AI official website ↗︎
Finally, let’s talk about an incredibly powerful—and FREE—method of upscaling your art: with the help of a vast number of AI models trained with deep learning methods to reconstruct high-quality images from less detailed originals. There are literally hundreds of openly available models, trained on thousands of various subjects. There are more specific (e.g., trained on video game screenshots, or designed to restore old photographs), and more general ones (for instance, general upscalers, or image sharpeners).
With so many of those models, the question is: how do I apply them to my images. This is where it might get tricky. There are simply too many practical options varying in level of complexity, comfort, and control.
Some can be experimented with in Google Colab notebooks↗︎, allowing you to run AI models in the cloud without local setup. But such options require a certain technical savvyness, or some level of patience to learn how they work, if you are a total beginner. Some are available on a model’s website, like Real-ESRGAN XL↗︎. However, prepare for a possible wait—a downside that many free tools have. With Real-ESRGAN XL playground, I stopped counting after 10 minutes of waiting. :)
A way for the technically-advanced readers of this guide is seting up a Python environment on your computer and install libraries like those offered by TensorFlow↗︎ or PyTorch↗︎ to run models locally.
And there are also websites like Hugging Face↗︎ that provide accessible interfaces to various pre-trained models. This is one of the easiest way to use different super-resolution models to enhance your images without deep technical know-how (although this might be much slower or have less options/less control over what you can achieve otherwise). Still, there are some quite powerful options, like Faizan141's Image Upscaler↗︎.
But in this guide, I will share a method that I found the most efficient for myself and that, I believe, strikes a “golden middle” between convenience and the effort/knowledge required.
For this test I picked the famous Real-ESRGAN as my test model. And here is the second part of that combination: chaiNNer.
chaiNNer is a node-based image processing GUI. It’s an app designed for easy customization and ”chaining“ of image processing tasks. Originally developed for AI upscaling, it has now evolved into a versatile and powerful tool for programmatic image processing.
With it, you can design and run complex tasks—handle intricate image processing workflows—through simple (and visual!) node connections. For instance, you can design a workflow to upload a batch of images, rename them in a specific way, upscale them with a select model and pre-set parametres, apply color correction, crop them, and sharpen the result with another AI model. And then save them with various presets to different folders. There are dozens of tools available in chaiNNer as nodes.
But for my setup, I just need four—with self-explanatory names: Load Image, Load Model, Upscale Image, Save Image.
Each node has a set of simple setting, and color-coded output and input ”dots.“ By dragging the “dots” you can connect output of one node to the input of another. Load Image’s yellow output connects into Upscale Image’s input.
The number of such steps/nodes/connection is unlimited. In our case, the model is connected, i.e. applied to the same Upscale Image node that provides the final result via connection to Save Image’s input. Just press the Play icon on top of the screen, et voila!
Incredibly versatile tool with infinite number of possible chains for any need imaginable—going way beyond just upscaling. Learn more at chaiNNer official website↗︎
There is a simply wonderful free catalog of available models—OpenModelDB↗︎. It contains searchable information about hundreds of downloadable models with details, and Before-After samples—similar to what we have in this guide—that you can filter by the subjects they were trained on.
We will continue reviewing different upscalers, both stand-alone apps and online apps. Our next guide will be devoted to “creative upscalers”: Leonardo AI, Magnific AI, and Clarity AI. Stay tuned!
Happy midjourneys,
—Andrei
If you're considering acquiring any upscaler from this guide, please, consider doing so using the links in the article. By doing so, not only will you be purchasing an excellent upscaler, but also supporting Midlibrary in the process!
You can help us maintain and expand Midlibrary and produce more regular educational content of higher quality. And keep it free for all!
Midlibrary Catalog grows largely through the contributions of our Community.
Thank you for taking time to share your suggestion!
We do our best to keep this website running as smoothly as possible.
However, stuff happens. Thank you for letting us know about it!
Every week we publish a new Midjourney study and a new Editor's Pick.
Receive our newsletter to never miss an important Midlibrary update!