The Adobe Firefly model, which allows you to generate images, will be enriched with several important functions, including the possibility of importing an image for inspiration. Ultra-detailed prompts are a thing of the past.
Launched in March 2023, Adobe Firefly artificial intelligence has already generated more than seven billion images. A number that is all the more impressive given that 90% of its users are not used to Adobe tools, which potentially creates a new market for the American company. Adobe Firefly is available on its own site (as a competitor to Midjourney or Dall-E) or directly within Adobe software, such as Photoshop.
On April 23, during its MAX London conference, Adobe announced several new features to come for its model, now capable of imitating an image (no need to prompt, just send him a photo to reproduce). Little by little, Adobe is positioning itself as a key player in the generative AI revolution.
Reference images: the future of generative AI?
Have you ever tried to generate an image with Midjourney, Dall-E or Firefly, without achieving the expected result? No matter the prompts, the end result does not always look like what we hope.
With the “Reference Image” function, Adobe Photoshop solves the problem in an interesting way. Want to put a blue suit and yellow tie on someone? No need to describe it. Find a photo of the famous costume, add it to the “reference image” submenu, circle the area to modify and click on generate. Photoshop will add a copy of the costume to the real photo. See below.
In addition to the reference image, the user can use the prompt window to refine their instructions. The idea is that Photoshop does not stupidly copy the image given to it, but takes inspiration from it. The result can sometimes remain approximate (you cannot copy a face for example), but a design can be copied in broad outline. Numerama has not yet been able to try the tool, but we wonder what will happen to branded products. Could a Nike logo on a shoe be automatically imitated by Firefly, for example?
In addition to the reference image, Adobe Firefly, in its web version, gains a “reference structure” system. It is intended for images created from scratch. The idea is to upload a drawing with lines and shapes so that the AI generates an image conforming to the locations imagined by its user. Which, once again, simplifies the exercise of the prompt.
Other new features of Photoshop and Firefly
In addition to reference images, Adobe is planning other new features to improve Firefly and Photoshop. Among them :
- The ability to “prompt” an image directly into Photoshop, to create an image from scratch.
- The ability to generate a variation of an image, with several suggested choices.
- The ability to add a background to an image, automatically isolating the subject from a monotonous background.
- A Firefly gallery to suggest prompt ideas.
- An improved model, with more details and more realism.
- Better suggestions in Firefly, to help refine a prompt.
By default, Creative Cloud subscribers are entitled to 1,000 uses of Firefly per month (other plans allow you to purchase more). These new features will be deployed today on Firefly Web and Photoshop beta, pending a stable version deployment in the coming weeks. Adobe says it wants “eliminate certain repetitive tasks so that its users go faster”which opens the door to other developments in the coming months.
Subscribe to Numerama on Google News so you don’t miss any news!