Can’t Think of the Perfect Gen AI Image Prompt? Adobe Says It Can Help – CNET

If you’ve ever been frustrated trying to come up with the exact right text prompt for a generative AI tool to create an image that matches a scene you have in your head, Adobe says it has a solution.

On Tuesday, the software giant announced it is adding a feature called Structure Reference to its generative AI tool Firefly to “bring users a new level of creative control” when using words to generate images.


A little over a year into the generative AI boom, tech companies like Adobe, OpenAI, Google, Microsoft and Anthrophic are increasingly focused on image and video capabilities as they vie for relevancy in a sector projected to reach $1.3 trillion in revenue by 2032. And with options like Dall-E 3, Google Gemini and Midjourney, as well as OpenAI’s Sora video generator (not yet in wide release), the competition is stiff.

Like Firefly’s Style Reference feature, which taps into users’ preexisting images for aesthetic inspiration, Structure Reference allows users to add their own images to generate variations with the same or similar layouts. According to Adobe, this makes it easier for users to generate what they’re envisioning without having to write the perfect prompt.

Structure Reference is available as of Tuesday. Style Reference launched in October.

Jonathan Pimento, group product manager of gen AI at Adobe, said to think of Structure Reference as where and how the objects are placed within an image, while Style Reference is more about, well, style.

“Structure is the composition of your image, and the Style is the look and feel of it,” he added.

Users can combine the Style and Structure Reference features and use a slider to tell Firefly how much each feature should influence the resulting images.

Applications include redesigning a room based on an existing picture, bringing a child’s drawing to life and colorizing black and white images.

One potential advantage to Firefly is that it has been trained on Adobe’s stock image collection, which Alexandru Costin, vice president of generative AI at Adobe, said includes nearly 400 million images. That’s as opposed to data from the Internet, which can pose problems like copyright issues.

“We use that curated dataset to train models that cannot generate IP or trademark or recognizable characters because it has never seen it,” he added.

According to Adobe, Firefly users have generated more than 6.5 billion images since the tool’s debut in 2023. Since then, Adobe has embedded Firefly capabilities into its Creative Cloud applications, including Photoshop.

“We’ve been investing in [generative AI] because we understood how much this will change creativity,” Costin said.

Editors’ note: CNET is using an AI engine to help create some stories. For more, see this post


Leave a Reply