Exciting news 🎉🎉🎉 for Photoshop users! Adobe has integrated a range of Firefly-based features into the software, bringing the power of generative AI to image editing. These new additions allow users to extend images beyond their borders with Firefly-generated backgrounds, add objects to images using generative AI, and utilize a new generative fill feature for precise object removal, surpassing the capabilities of the previous content-aware fill.
At present, these features are only available in the beta version of Photoshop. However, Adobe has also made certain capabilities accessible to Firefly beta users on the web. It’s worth mentioning that Firefly users have already created over 100 million images on the platform.
One fascinating aspect of this integration is that Photoshop users can now utilize natural language text prompts to describe the type of image or object they want Firefly to generate. As with any generative AI tool, the results can sometimes be unpredictable. By default, Adobe provides users with three variations for each prompt, although unlike the Firefly web app, there is currently no option to iterate on a given result and view similar variations.
|Image Credits: Adobe
To achieve this functionality, Photoshop sends portions of an image to Firefly—though not the entire image, although Adobe is also experimenting with that approach—and creates a new layer for the generated results.
Maria Yap, the Vice President of Digital Imaging at Adobe, provided a demonstration of these new features ahead of the official announcement. As is typical with generative AI, it can be challenging to anticipate the model’s output. Nevertheless, some of the results were surprisingly impressive. For instance, when asked to generate a puddle beneath a running corgi, Firefly took into account the overall lighting of the image, even producing a realistic reflection. While not every result was flawless—a bright purple puddle was one option—the model generally excelled at adding objects and extending images beyond their original frames.
Given that Firefly was trained on Adobe Stock photos and other commercially safe images, it is particularly adept at landscapes. However, like most generative image generators, it struggles with text.
Image Credits: Adobe
Adobe has also implemented safety measures to ensure the model’s results remain appropriate. This includes the use of a carefully curated training set and additional prompt engineering techniques. Certain terms and words that are deemed unsafe are excluded. Furthermore, Adobe is considering an additional hierarchy of precautions. For example, if a user selects an area with a significant amount of skin, Adobe may choose not to expand the prompt in order to maintain predictability and user comfort.
As with all Firefly images, Adobe automatically applies its Content Credentials to any images that utilize these AI-based features, ensuring safe and responsible usage.
Many of these features would also be highly valuable in Lightroom. Yap acknowledged this and, while not providing a specific timeline, confirmed that Adobe has plans to bring Firefly to its photo management tool in the future. Stay tuned for further updates!
DISCLAIMER: please note that any download or usage of software introduced in the aforementioned article is solely at the user’s discretion and responsibility. The website and its creators bear no responsibility for any consequences that may arise from downloading or using the software. It is recommended to carefully read the terms and conditions before downloading or using any software.