Currently in beta, Adobe Firefly is billed as the ultimate AI solution for creators, allowing them to bring whatever is in their heads to life.
One of the biggest pitfalls of AI image generators like Midjourney and DALL-E is that they cannot handle text. If you need a logo for your business or a graphic, this can make it somewhat limited. While you can create logos and other designs, they usually require minor editing in Photoshop. Adobe aims to change this.
The first Firefly model creates high-quality images and text effects in the same way that other generators do.
The difference here is that Firefly will be trained on different models to generate a more diverse range of images.
There is also a text-effect feature allowing users to create text in an array of different patterns and colors such as flames or animals, though this is fairly basic for now.
These tools will be integrated across Creative Cloud, Document Cloud, Experience Cloud, and Adobe Express following a trend that is also seeing Microsoft implement AI tools across their products in Microsoft Co-pilot and Google add text to image and AI text generation to their Google Work space. An AI tool for recoloring vectors in Illustrator became available yesterday.
The first version of Adobe Firefly, however, is just the start of Adobe’s plan. Adobe wants to include video editing via a text prompt, eg change the weather or landscape in this scene.
We’ve already seen a basic, heavily stylized version of this in the text-to-music video generator Kaiber AI. Adobe’s plans also include the ability to “make unique posters, banners, social posts, and more with a simple text prompt.”
Transparency and Consent Policy
Adobe leans heavily on transparency and consents in its approach to AI. In response to criticism that leveraging artists’ images for training purposes is a violation of copyright and even theft, Adobe will introduce a tag called “Do Not Train”.
Users who flag their work with the tag will not have their content used in model training. Conversely, Adobe wants to empower creators to be able to train their models with their own content, creating a unique style or brand that only they can generate.
Adobe has also allowed users to sell and license AI-generated images on its Stock image library, so long as they are labeled as AI generate for transparency.
The pricing structure is currently unclear as the tool is still in beta. Users must apply to join the beta.
However, it’s expected to form part of Adobe’s Creative Cloud subscription package.