3 min read
Adobe introduced its new ‘Firefly’ image generator that takes art from its stock library and public domain. The main aim is to compensate stock image creators eventually.
According to Adobe, the company has used millions of assets from its stock service to train its new AI.
On Tuesday, Adobe launched its generative AI model family called ‘Firefly’. As per the company, it’s not like the other scrappy AI image generators. Instead, it behaves like young people often do during puberty.
Adobe claimed it created its service exclusively using images from its stock image site, openly licensed work, and public domain content rather than training data collected online.
The service is initially available online, taking the standard text-to-image diffusion AI model with some slight changes.
How Does It Work?
Users input a prompt as usual but, as per the company, a user can modify the image afterwards with a different aspect ratio, art style or even lightning.
The company has made this AI with great further plans. At the same time, the most exciting aspect of this innovative AI is its training data.
According to Adobe, “the current model is trained on a dataset of Adobe stock, along with openly licensed work and public domain content where the copyright has expired”.
Adobe’s VP of generative AI, Alexandru Costin, told the news, “their system can’t generate content like other brands or IPs simply because it has never seen the brand content or trademarks”.
Adobe claims that it was developing a “compensation model” for contributors. Adobe told Gizmodo that the company would reveal more advanced features soon.
Adobe is one of the famous image creation software. Despite being one of the industry’s most widely used stock image services, Adobe has been conspicuously silent on the AI image debate.
Although shutter stock debuted its own AI image generator based on OpenAI’s Dall-E earlier this year, stock image service.
Shutterstock also claimed that “it plans to compensate artists whose images were sued to train the AI.
Whereas the opposite approach has been taken by Getty Images, which has outlawed all AI-generated images and even filed a lawsuit against stable diffusion creators, accusing them of stealing its copyrighted images to use as training data.
According to Gizmodo, Firefly was trained on more than 330 million assets from Adobe Stock and additional millions from outside sources. In contrast, the LAION-5B training set contains more than 5.85 billion images, their textual metadata, and the data used for Stable Diffusion using scraped web images.
However, users can work directly with these innovative AI systems within its creative cloud and document cloud software.
Adobe has several plans to incorporate custom vectors, brushes and textures usable in Adobe’s other suites of programs like Illustrator and Photoshop.
Alas, Adobe says, users can change video images in massive ways, even changing the “mood” or the weather of an image from summer to winter with a single prompt.
Read more:
ChatGPT-4 Language Model Can Handle Text And Images
Google Will Soon Blur Explicit Images in Search Results
