Bring Your AI Art To Life – Generating, Upscaling, Copyright & Printing
Here at Bumblejax, we work with many talented and hard-working artists who use our service to create gallery-quality wall art. The artists we work with include everyone from portrait photographers, to digital artists to painters to interior designers to so many creatives in between a simple label.
You’ve no doubt been hearing more and more about artificial intelligence (AI) tools, particularly in the past few months to create articles, software programs, video games, and yes.. art. No doubt this has created lots of controversy and will undoubtedly change the landscape across industries forever. We’ve only begun to scratch the surface here of what is possible in AI generated art, the copyright issues around it and how it changes how artists create art and consumers buy it.
One thing is for certain. It’s helped fuel the creativity of many who are mixing and matching different styles, subjects, genres, and colors to produce art that speaks to them and is personalized to their exact taste. It’s allowed artists to create AI models based on original works to produce residual works. Interior designers can take photos of a space and produce many variations and styles to help the client design their perfect space. The opportunities are endless across many industries.
What Is AI Art? How Is It Generated?
AI art, also known as artificial intelligence art, refers to the creation of visual artworks, designs, or compositions using algorithms, machine learning, and artificial intelligence technologies. AI art often involves the use of deep learning techniques, generative networks and/or neural networks to analyze and learn from existing artistic styles, patterns, and compositions, and then generate new, unique, and visually appealing artworks based on these learnings.
AI art can involve various forms, including:
- Generative art: AI algorithms create new, unique visual compositions based on input data, such as a collection of images or text. The AI “learns” from the input data and generates new artworks that reflect the patterns, styles, or themes found in the input data.
- Style transfer: AI algorithms take an existing image and apply the artistic style of another image to it, creating a new image that combines the content of the first with the style of the second. This process often involves the use of convolutional neural networks (CNNs).
- Interactive AI art: AI algorithms are used in real-time to generate visuals or modify images based on user input, such as gestures, sounds, or other data. This can create immersive, dynamic experiences in which the audience can directly influence the artwork.
AI art challenges the traditional boundaries of artistic creation and explores the intersection of art, technology, and human creativity. It raises philosophical questions about the role of the artist, the nature of creativity, and the evolving relationship between humans and machines in the creative process.
Recently we’ve seen an uptick in customers wanting to print AI generated art so we wanted to discuss ways to create it and tips for printing it for optimal results. It should be noted that the output of these AI image generators is fairly low resolution so they require upscaling with … another AI tool! That’s right .. upscaling images has gotten significantly better in recent years thanks to the power of AI. Some tools are better than others and one tool might be great for one image, but not as good for another. Then there are other improvements that can be made such as color enhancement, denoising, sharpening, lightening, etc.
Best AI Art Generators
There are several websites and platforms that allow you to create AI-generated art using a variety of techniques, like style transfer, generative adversarial networks (GANs), and deep learning models. Some of the popular websites for creating AI-generated art include:
- Bing Image Creator: Easy to access and free. Runs on DALL-E. Outputs at 1024x1024px.
- Stable Diffusion: Free and no signup required so a great intro to text to image AI generation. They even have a database where you can search for prompts by other users to get ideas. Only outputs at the very low resolution of 512x512px.
- DALL-E 2: One of the most popular image creators at the moment is DALL-E 2 which is a neural network developed by OpenAI (the same company that developed ChatGPT) that can generate images from textual descriptions. You will need an OpenAI account to play with it and purchase credits for continued use. Highest output at 1024x1024px.
- Midjourney: Paid subscription only, no free trial. Outputs the highest resolution images at 2k+ pixels, but incredibly clunky and confusing signup process (requires a Discord account). We recommend using Bing or DALL-E 2 and just using an AI resizing tool to get high res images. (Note: I went through the entire process to create an image and by the time I got to the “you need to subscribe” link I gave up). This is a service that would have to be significantly better than Bing/Dalle/Stablefusion to consider paying for it. Plans start at $10/month.
- Dream by Wombo. Both an online app and iPhone app with thousands of positive reviews, Wombo is an easy to use AI art generator. Just type a prompt, pick a style, and get the results. There are paid features as well.
- RunwayML: RunwayML is a user-friendly platform that allows artists and designers to create AI-generated art through an intuitive interface. Mostly focused on video but provides image editing and creation tools as well. You can experiment with different machine learning models, including StyleGAN, BigGAN, and more.
- Artbreeder: Artbreeder allows you to create AI-generated art by blending and evolving images. You can upload images, mix them with others, and generate new, unique pieces of art to share with the community.
Please note that the availability of these platforms may change over time, and new tools may emerge as AI-generated art continues to evolve.
Best Bet? Try them all and see what you like the best! Depending on the style, subject matter, etc one image generator might be better than another. Just look at the sample we did for “flying monkey eating a banana as Basquiat painting” across Bing, Dalle & Stable. In this example, Bing is the clear winner in my opinion. Vastly superior! That may not be the case for other queries.
Upscaling Your AI Art For Printing
So you’ve played around with the AI art generators a bit and come up with a cool image that you’d like to get printed. The big problem? The output of these systems is very low – typically 500-1000px. Generally speaking, you need about 1000 pixels for every 10″ of print. Unless you want a 5-10″ print you’ll need much better files. This is where AI comes in again.
There are now many AI resizing tools out there that will artificially add pixels so that you can print large. The quality of these tools varies greatly so you’ll want to try them all out and see what you like best. Here at Bumblejax, we use a combination of AI tools (one tool might work better on one image vs another) and lots of experience optimizing files for print. There’s a lot more that goes into getting an image to print large other than simply upscaling. There are denoising, sharpening, lightening and even color improvements that can be made to make sure your image is the best it can be to go big. All that said, Topaz Labs GigaPixel tool comes up often as a top tool and it’s one we definitely have in our arsenal. It doesn’t come cheap at $99/year, but they do have a free trial for 30 days so give it a shot. If you want an expert to take care of it for you and get the best image possible we offer complete enhancement services for $25-$50/per file. We send a before and after pic so you know exactly what you’ll get before paying. Feel free to reach out for more details.
SAMPLE AI Image Enhancement
In this example, I generated an image of a flying monkey eating a banana as a Basquiat-style painting and I have to admit I kind of like it! The goal was to get it to work by enhancing the original low res image from Bing in order to print it fairly large (in this case 36×36″). As you can see in the before & after sliding image below the results are miraculous. I’d be comfortable printing this image all the way up to 42″ square! The original image could only be printed up to about 12″ square so a more than 3x print size improvement after upscaling, denoising, and sharpening.
Image Creator: Bing
Prompt: “flying monkey eating banana as basquiat painting”
Output: 1024x1024px
Print Size: 36×36″
Enhancements: Upscale, Denoise, Sharpen
AI Art Copyright
The rules around AI-generated art and copyright are complex and evolving quickly. It’s essential to consult with a legal professional familiar with the specific jurisdiction to ensure compliance with the most up to date copyright laws and regulations.
Where it stands right now? On March 16th, 2023 the US Copyright Office issued a statement clarifying its position on AI generated art. Here’s the gist and you can read the full statement here.
- “Non-human” works of art are not eligible for copyright
- Using AI generated images within human made art may be eligible (ie. combining AI images into original art or using them within a painting or mixed media
- Partial protections will be common where the human generated portion will be protected while the AI generated portion will not. For example a comic book was submitted for review and ultimately the AI generated images for the book were not copyrighted but the human generated writing and other elements were protected
Perhaps the biggest question and one that continues to remain shrouded in uncertainty are the AI-generated images based on the style of original works. Artists are upset and claim that images generated that emulate their “style” breach copyright while others (the companies producing these images) say it doesn’t breach copyright law. A class action lawsuit was filed against DeviantArt and two AI art companies over their use of copyrighted images to train artificial intelligence tools. They claim that the generators downloaded and used billions of copyrighted images without obtaining the consent of or compensating any of the artists.
Printing AI Art For Best Results
AI-generated art can be a fascinating fusion of creativity and technology. To ensure the highest quality when printing AI art, consider the following:
- High-resolution images (see above): AI-generated images should have a high resolution to ensure crisp, clear prints. If you need to upscale your image (and currently you will need to!), you can use AI-based tools like Let’s Enhance, Topaz Gigapixel Or ON1 Resize to improve the image quality. You can also reach out to us to professionally optimize your image!
- Select the right printing method and product that best suits your needs and the art style. When customers come to us with an image and don’t know where to start we also start with two questions to hone in on the product or two that works best for them. – What look do you prefer? Softer/matte or glossy vibrant? – What is your budget? Do you want the best of the best or have a tight budget you need to stay within? We also send them over to our detailed buy guide which guides them through all of our products. Considering we offer acrylic prints, metal prints, canvas, bamboo, birch, dibond and foamcore it’s a lot to consider! We’re available via email, phone, text and live chat as well to discuss your printing needs.
- Color management: Make sure your printer, monitor, and other devices are properly calibrated to accurately represent the colors of your AI art. Use ICC profiles and color management tools to maintain color consistency across devices. Here at Bumblejax, we are constantly calibrating our systems to ensure accurate color so if your monitor is properly calibrated, the colors should be a close match. That said, there is never any substitute for a test print. If you’re highly concerned about matching the color you see on your monitor with a print from us or someone else, we recommend a small test print especially if the colors are out of gamut. If you don’t know what gamut is or how to test for it we can help. It just means that there is a much higher chance that the color you see on your monitor won’t match the print output.