Adobe Stock, one of the biggest stock photography and imagery collections, is getting new AI updates. Adobe's AI model Firefly is powering two new features: generative edits and generative variations. They do exactly what they sound like; with generative edits, you can use AI to tweak existing images, and generative variations use stock photos as reference guides to create new images.
Generative edits will include the ability to remove a stock image's background, replace a background with AI-generated elements and expand an image, like turning a square image into a 16:9 rectangle. These tools use the base stock image to create a cohesive image with the original and AI-generated elements.
Generative variations will create entirely new images, using the stock image as a style or composition reference. So if you like the set-up of one image -- the elements' placement, depth, clarity, etc. -- you can tell Firefly to reference that design when generating whatever you prompt it with. You can do the same thing with style, guiding the AI toward a particular color palette or aesthetic.
Photographers who submit their work to Adobe Stock already agree to let the company train its AI models on their creations, and they won't be able to opt out of these new AI tools being used on their work. Stock contributors will be paid for times when their work is used to generate alternative versions and those AI versions are downloaded. Adobe said that adding these tools "increases the versatility of an asset, making it more likely to be licensed."
Read More:Â Can Adobe Turn Its Creators From AI Skeptics Into Believers?
Adobe has been investing heavily in AI. Its photo and video editing programs recently got a variety of new AI-powered editing tools, and soon, Adobe users will be able to generate video in addition to images with its Firefly model. The company is also investing in better AI labels with content credentials and the Content Authenticity initiative.
Stock libraries like Adobe's have become even more important in the age of AI as tech companies look for sources of content to train their models on. Adobe's AI models are only trained on Adobe Stock and other public domain content (like content where the copyright has expired), not from individual creators' content or pulled from the open web.