Adobe launches Firefly: a new suite of text-to-image generative AI tools following in the footsteps of DALL-E and MidJourney. To set itself apart from the competition, the American giant intends to emphasize respect for artists and intellectual property… find out all you need to know!
For several months now, “text-to-image” artificial intelligence tools such as DALL-E and MidJourney have been encountering a viral success.
These AIs make it possible to create an image from a simple prompt textual prompt. They are used by graphic designers, architects and many artists.
During the summer of 2022, a painting created by MidJourney won an art competition in Colorado. Recently, the first manga created with this tool has been published, and Coca-Cola is organizing a competition to design the new label for its soda with DALL-E.
However, these artificial intelligence solutions are also controversial. They have been trained on many artists’ works without asking for their consent, and many feel they are simply victims of plagiarism.
Now, it’s Adobe’s turn to launch its own AI image generator: Firefly. And to set itself apart from the competition, the American giant is counting on assert its respect for creators…
What is Adobe Firefly?
To be exact, Adobe Firefly is not a single tool, but a ” family of generative AI models “.
For the time being first two tools have just been launched by the company. The first works like DALL-E and MidJourneyby creating an image from a text prompt.
The second model generates stylized text. It could be defined as an AI-based variant of WordArt.
These two tools are accessible via web browserand generate JPEG or PNG images in 2000×2000 resolution. They are aimed at illustrators, conceptual artists, graphic designers and other 2D artists.
In the future, Adobe also plans to offer Prompt-based video editing text editing and 3D capabilities.
Adobe’s entry into the field of generative AI
This launch is important for Adobewhich has long held a central position in the creative applications ecosystem.
Over the past year, the company has remained in the shadow of new players offering powerful tools for image, video and sound creation.
However, according to Adobe’s Vice President of Generative AI, Alexandru Costin, the company ” is not afraid of change, and embraces it “.
She counts ” bring these capabilities to our products so that customers don’t need to know what’s generative and what isn’t “.
AI that respects intellectual property
While many generative AIs already exist, Adobe is one of the few companies willing to demonstrate data transparency on which its models are trained.
According to the company, all the data used to feed its AI are royalty-free, licensed for training, or taken from the Adobe Stock database for which it holds the rights of use.
As a result, Adobe’s offer should be much less frustrating for artists, and more attractive to them. adapted for corporate use.
As Costin explains, this AI can generate high-quality content and does not risk recreating intellectual property of other brands, since it has never seen it.
Also, Adobe plans to pay artists who contribute training data. This partnership will not be offered at launch, but this “compensation strategy” will be developed before the end of the beta.
Another planned feature will allow artists to train the system on their own work so that it can act as an assistant, generating content in their own Photoshop style.
To avoid drift, Adobe is thinking of a way prevent users from training AI on the work of others artist. It would be possible, for example, to compare images uploaded to its Behance art-sharing social network.
An artist will also be able to block AI training on his or her work. The company develops a “Do Not Train” system (do not train) system to be incorporated to image metadata. Discussions are underway with other model creators to generalize this approach.
This idea is reminiscent of the ” Have I Been Trained “launched by Stability AI to enable artists to check whether their works have been used to drive the Stable Diffusion image generator and request their removal from the dataset.
A more intuitive tool than its competitors
Adobe’s ambition is also to offer easier-to-use AI tools than the competition. Rather than having to compose a complex line of descriptions for an image’s style, users will be able to exploit pre-built styles, lighting and aspect ratios.
It will even be possible to apply these effects to a previously generated imagerather than having to generate a new image for each modification.
Similarly, the textual stylization tool offers options for font size or background color.
Artificial intelligence soon to be incorporated into Adobe tools
Adobe is already planning to integrate Firefly’s AI into its creative applications. In particular, it will be possible to generate “outpainting” for Photoshop. On Premiere, the AI will be able to modify the style or image colors with a simple prompt.
On Illustrator, it can generate vector variations for hand-drawn sketches. It is not yet known when these new features will be deployed.
Adobe Firefly: price and availability
For now, Adobe Firefly is available in beta version. The suite of tools is only accessible via a web site at this address.
It is compatible with browsers Google Chrome, Safari and Microsoft Edge. Smartphones and tablets, however, are not yet supported.
It is you don’t need a Creative Cloud account account to request access. However, the number of users will be voluntarily limited by Adobe.
In parallel, a new package for businesses has just been launched for Adobe Express. This subscription will provide access not only to Adobe Firefly, but also to images, videos, stock fonts and creative templates. Pricing has not yet been announced.
In the future, the firm plans to integrate its generative AI tools with its suite of creative applications including Photoshop, Illustrator and Premiere…