Adobe has launched two new versions of its text-to-image generative AI model alongside a host of new Firefly features and Creative Cloud app updates coming to Photoshop and Illustrator.
The fourth-generation Firefly Image models follow a similar precedent that OpenAI and Google have set for their chatbot releases, providing users with a choice between one model that’s suited for speed and efficiency, and another for more demanding tasks.
Adobe says that Firefly Image Model 4 is its “fastest, most controllable, and most realistic Firefly image model yet,” allowing users to generate images in up to 2K resolution with more control over style, format sizes, and camera angles. The updates from its predecessor are designed to improve the quality of image outputs while allowing them to be generated “quickly and efficiently.” For image prompts that require more “detail and realism,” Adobe is also launching Firefly Image Model 4 Ultra, which is more capable of rendering “complex scenes with small structures.”
The new Firefly image models are generally available now via the Firefly web app alongside Adobe’s text-to-video and text-to-vector models, which previously launched in public beta. A brand new Firefly web platform tool that’s also launching in public beta today is Firefly Boards — a FigJam-like collaborative generative AI moodboarding app that was introduced during Adobe’s Max event in October as “Project Concept.” Adobe also says that a Firefly mobile app is “coming soon” for iOS and Android devices.
The Firefly web app now also gives users access to third-party AI models when generating images or video. Users can select between OpenAI’s new GPT image model or Google’s Imagen 3 for images, or Google’s Veo 2 model for video alongside Adobe’s own AI models, with support for Luma, Pika, Runway, fal.ai, and Ideogram “coming soon,” according to Adobe.
The company describes these third-party offerings as being available for “experimentation” rather than publishable work, however, and clearly marks its own models as being “commercially safe.” That’s not terribly surprising given Adobe trains its AI models on public or licenced content, while OpenAI, Google, and Runway can’t claim the same.
