The Firefly tools begin with image creation and font styling but will spread to Photoshop, Illustrator and other software.
In 2022, OpenAI's Dall-E service wowed the world with the ability to turn text prompts into images. Now Adobe has built its own version of this generative AI technology with tools that begin a technological overhaul of the company's widely used creative tools.
On Tuesday, Adobe released the first two members of its new Firefly collection of generative AI tools for beta testing. The first tool creates an image based on a text prompt like "fierce alligator leaping out of the water during a lightning storm," with hundreds of styles that can tweak results. The other applies prompt-based styles to text, letting people create letters that look hairy, scaly, mossy or however else they want.
Firefly for now is available on Adobe's website, but the company will build generative AI directly into other tools, starting with its Photoshop image editing software, Illustrator for designs and Adobe Express for creating quick videos. The company hasn't revealed its pricing approach for the new tools.
Creative professionals might see Firefly as an incursion into their creative domain, going beyond mechanical tools like selecting colors and trimming videos into the heart and soul of their jobs. With AI showing new smarts when it comes to translating documents, interpreting tax code, composing music and creating travel itineraries, it's not irrational for professionals to feel spooked.
Like other AI fans, though, Adobe sees artificial intelligence as the latest digital tool to amplify what humans can do. For example, Firefly eventually could let people use Adobe tools to tailor designs to individuals instead of just creating one design for a broad audience, said Alexandru Costin, vice president of Adobe's generative AI work.
"We don't think AI will replace creative creators. We think that creators using AI will be more competitive than creators not using AI. This is why we want to bring AI to the fingertips of all our user base," Costin said. "The only way to succeed in AI is to embrace it."
Adobe's Firefly products are trained from the company's own library of stock images, along with public domain and licensed works. The company has worked to reduce the bias in training data that AI models can reflect, for example that business executives are male.
Artificial intelligence uses processes inspired by human brains for computing tasks, trained to recognize patterns in complex real-world data instead of following traditional and rigid if-this-then-that programming. With advances in AI hardware, software, algorithms and training data, the field is advancing rapidly and touching just about every corner of tech.
The latest flavor of the technology, generative AI, can create new material on its own. The best known example, ChatGPT, can write software, hold conversations and compose poetry. Microsoft is employing ChatGPT's technology foundation, GPT-4, to boost Bing search results, offer email writing tips and help build presentations
AI tools are sprouting up all over. Adobe has used AI for years under its Sensei brand for features like recognizing human subjects in Lightroom photos and transcribing speech into text in Premiere Pro videos. EbSynth applies a photo's style to a video, HueMint creates color palettes and LeiaPix converts 2D photos into 3D scenes.
But it's the new generative AI that brings new creative possibilities to digital art and design.
"It's a sea change," said Forrester analyst David Truog.
Alpaca offers a Photoshop plug-in to generate art, and Aug X Labs can turn a text prompt into a video. Google's MusicLM converts text to music, though it's not open to the public. Dall-E captured the internet's attention with its often fantastical imagery -- the name marries Pixar's WALL-E robot with the surrealist painter Salvador Dalí.
Related tools like Midjourney and Stability AI's Stable Diffusion spread the technology even further.
If Adobe didn't offer generative AI abilities, creative pros and artists would get them from somewhere else.
Indeed, Microsoft on Tuesday incorporated Dall-E technology with its Bing Image Creator service.
Training AIs isn't easy, but it's getting less difficult, at least for those who have a healthy budget. Chip designer Nvidia on Tuesday announced that Adobe is using its new H100 Hopper GPU to train Firefly models through a new service called Picasso. Other Picasso customers include photo licensing companies Getty Images and Shutterstock.
Developing good AI isn't just a technical matter. Adobe set up Firefly to sidestep legal and social problems that AI poses.
For example, three artists sued Stability AI and Midjourney in January over the use of their works in AI training data. They "seek to end this blatant and enormous infringement of their rights before their professions are eliminated by a computer program powered entirely by their hard work," their lawsuit said.
Getty Images also sued Stability AI, alleging that it "unlawfully copied and processed millions of images protected by copyright." It offers licenses to its enormous catalog of photos and other images for AI training, but Stability AI didn't license the images. Stability AI, DeviantArt and Midjourney didn't respond to requests for comment.
Adobe wants to assure artists that they needn't worry about such problems. There are no copyright problems, no brand logos, and no Mickey Mouse characters. "You don't want to infringe somebody else's copyright by mistake," Costin said.
The approach is smart, Truog said.
"What Adobe is doing with Firefly is strategically very similar to what Apple did by introducing the iTunes Music Store 20 years ago," he said. Back then, Napster music sharing showed demand for online music, but the recording industry lawsuits crushed the idea. "Apple jumped in and designed a service that let people access music online but legally, more easily, and in a way that compensated the content creators instead of just stealing from them."
Adobe also worked to counteract another problem that could make businesses leery, showing biased or stereotypical imagery.
And it's working to address some other prickly AI issues. For one thing, it's tagging Firefly images with metadata indicating it's been created by an AI system using a metadata system from the Content Authenticity Initiative the company helped found. The CAI's initial purpose was to write metadata into image files to help news publishers and others ensure photos were real, not faked or edited.
For another thing, Adobe is letting creators use CAI technology to incorporate a "do not train" label in their imagery or other digital files that instructs AI training systems to exclude it from training data.
Compliance with the system is voluntary, of course, as with the "robots.txt" file website operators can use to exclude their sites from search engine indexing.
"Some might decide to not respect it, and there's nothing we can do about that," Costin acknowledged. But other AI training companies have expressed some enthusiasm might welcome a system for respecting artists' wishes, he said.
It's now up to Adobe to convince creative pros that it's time to catch the AI wave.
"The introduction of digital creativity has increased the number of creative jobs, not decreased them, even if at the time it looked like a big threat," Costin said. "We think the same thing will happen with generative AI."
Editors' note: CNET is using an AI engine to create some personal finance explainers that are edited and fact-checked by our editors. For more, see this post.