A lot of the hype and concerns about generative AI have been about text. But the systems that can generate images are also developing rapidly and dramatically. Often these share a similar structure to text-based generative AI, but can also be weirder and lend themselves to fun creative pursuits.
Image generators have been trained on billions of images, allowing them to create works once reserved for painters and other artists.
In some cases, experts may not be able to tell the difference between an AI-generated image and a real photo (a situation that fosters dangerous misinformation campaigns in addition to fun creations). And these tools are changing the way creative professionals work.
Compared to products such as ChatGPT, image generation AI tools are less developed. You have to go through a few more hurdles and it may cost you some money. But if you’re interested in learning the tricks, now’s the perfect time to start.
AI Photoshop
Adobe recently added generative AI capabilities to the beta of its iconic graphics software, Photoshop, and creators on social networks like TikTok and Instagram are talking about the feature.
I have a fair amount of experience with Photoshop. When I tested a new feature called Generative Fill, I was impressed with how quickly and competently the AI performed a task that would have taken me at least an hour to do by myself. In less than 5 minutes, and with just a few clicks, I was able to use this feature to remove objects, add objects, and replace backgrounds.
(To try these tools for yourself, first sign up for a free trial of Adobe Creative Suite. Then install the new Adobe Photoshop Beta, which includes Generative Fill.)
After installing Photoshop Beta, import your photos and try the following tricks.
— To change the background, click the Object Selection icon (box with an arrow), and on the Selection menu, click Invert to select a background. Then click the Generate Fill box and enter your prompt. Or leave it blank and let Photoshop come up with a new background concept.
— To delete an object, use the Lasso tool. In the photo of my bike, I wanted to remove the tractor behind the fence in the background. I traced around the tractor, clicked the ‘generate fill’ box and hit ‘generate’ without entering any prompts. The software correctly removed the tractor and filled the background while leaving the fence intact.
(Note: If you find it difficult to install Photoshop, you can test Adobe AI more easily by visiting the Adobe Firefly website. There, open the Generative Fill tool, upload an image, click the Add tool You can trace around your subject by using the
The New York Times photo editors do not manipulate or alter photos or use artificial intelligence to generate images. But after testing Generative Her Phil, my first thought was that photo editors working in other areas, such as marketing, could quickly lose their jobs.
When I shared this theory with Ely Greenfield, chief technology officer at Adobe, he was optimistic that while this might make photo editing easier, humans would still need it. He said he was.
“I can make very pretty images with it, but to be honest, I still make boring images,” he said. “If you compare the content that artists create when they put it in their hands and the content that I create, their content is much more interesting because they know how to tell a story. .”
To be honest, what I’ve done with Generative Fill is a lot less exciting than what other people have posted on social media. Tweeting about AI, Lorenzo Green said, Posted a collage of famous album covers, such as Michael Jackson’s “Thriller” enhanced with generative fill and Adele’s “21”. The results were very interesting.
Other image generators
Tools like DALL-E and Midjourney can create entirely new images in seconds. These work similarly to chatbots. Enter the text prompt. The more specific, the better.
To create a quality prompt, start with the media you want to emulate, followed by a subject line and additional details. For example, typing “picture of a cat in a sweater in a bright room” into the DALL-E prompt box produces such a picture.
Owned by Open AI, the makers of ChatGPT, DALL-E was one of the first widely available and easy-to-use AI image generators. Get 115 credits for $15. You can generate a set of 4 images using 1 credit.
Another popular image generator, Midjourney, is still in development, so the user experience isn’t as polished. The service costs $10 a month and requires you to join Discord, another messaging app, so filling in the prompts can be a little complicated. Nevertheless, the project allows you to create high-quality, realistic images.
To use it, join the Discord and request an invite to the Midjourney server. After joining the server, type “/imagine” inside the chat box, after which you will be prompted. Typing in “/imagine a cartoon cover with a corgi in a ninja turtle costume” generated a compelling series of images.
Entering a basic request is fine, but some people have found obscure prompts that produce exceptional results (the tech blog Beebom has a list of examples). At Columbia University, Lance Weiler teaches students how to use AI, including Midjourney, to create artwork.
Remember that whatever tool you use, you are responsible for using this technology responsibly. Technologists warn that image generators could increase the spread of deepfakes and misinformation. However, the tool can also be used in positive and constructive ways, such as making family photos look better or brainstorming artistic concepts.
In case you’re wondering, the hilariously insane image at the beginning of this story was created by a human, not an AI, by illustrator Charles Desmarais.
