Introducing Firefly, Adobe’s answer to new image generators that use AI

Listen to the audio version of the article

It must not have been easy days in the upper floors of the Californian digital graphics giant Adobe. For over 10 years, the creators of Adobe Photoshop in the 1990s, the most famous image editing software, have made artificial intelligence functions available to users in their digital creativity products. The appearance of image-generating intelligence services that have conquered the limelight in a very short time will have experienced it as lese majesty. Let’s talk about names on everyone’s lips such as Midjourney, Dall-E and Stable Diffusion. The response came shortly after the opening of the Firefly beta last week. It’s a new family of generative AI creative templates focused on creating images and text effects. For now it is limited, it is only for non-commercial use but will soon be integrated into Adobe products. We tried it in preview. While we are writing, the classic text-to-image mode (you write in natural language and the system generates an image) and that of graphic effects applied to the text are available. The first impression is that of being faced with a high quality service, perhaps less surprising than Midjourney (which we tried and compared the same prompts) but cleaner. And with ample room for growth if associated with the numerous Adobe tools. But the feature that we liked most is that it does not copy the work of others. At least on paper. Firefly is trained on a proprietary dataset of Adobe Stock and other openly licensed or public domain content. You are not violating anyone’s copyright. Adobe is developing a compensation model and will share details when Firefly exits beta. And they’re trying to figure out how to reward creators who agree to train machine learning algorithms, excluding the use of work produced by other creators.

There are still no clear and shared indications on this front. In January, the creators of Midjourney, Stable Diffusion were taken to court by several artists on charges of copyright infringement. The photo agency Getty Images has also turned to lawyers against generative AI. The feeling is that it’s late, as they say the toothpaste has come out of the tube and getting it back won’t be easy. As some experts have observed, the more time passes, the more difficult it will be to adjust these tools that learn from our questions. A few days ago Midjourney, the platform that generated the fake photos of the Pope dressed as a rapper, announced that it will stop offering a free trial period to users who want to test it because the function has recently been widely abused. Until now, Midjourney allowed anyone to generate 25 images for free before starting to pay a 10 euro monthly subscription. As mentioned, the toothpaste is almost completely out.