Adobe has launched two novel versions of its generative model AI text for the image along with many novel Firefly functions and updates of the Imaginative Cloud application in Photoshop and Illustrator.
The fourth generation Firefly Firefly Pictures models are in line with a similar precedent, which Openai and Google have determined for their publishing houses Chatbot, providing users with a choice between one model that is suitable for speed and performance, and the other for more demanding tasks.
Adobe says that Firefly Image Model 4 is “the fastest, most controlled and most realistic Firefly image model”, enabling users to generate images in a resolution up to 2K with more control over style, sizes of camera formats and corners. The updates from its predecessor were designed to improve the quality of image outputs, while allowing them to generate “fast and efficient”. In the case of image hints that require more “details and realism”, Adobe also launches the Firefly Image Model 4 Ultra, which is more capable of rendering “complex scenes with small structures”.
The novel Firefly painting models are now generally available via the Firefly internet application with the Adobe Text-Video model and the vector text that previously launched in the public version of beta. A novel tool for the Firefly internet platform, which is also introduced in the public version of beta, are Firefly boards-a similar to Figam, generative generative application AI moodboarding, which was introduced during the Adobe Max event in October as “Project concept”. Adobe also says that the Firefly mobile application “soon” on iOS and Android devices.
The Firefly internet application now gives users access to AI models other companies when generating images or video. Users can choose between the novel OPENAI or Imagen 3 Google GPT image model for photos or the Google Veo 2 for video model with their own AI Adobe models, with Luma, Pika, Runway, Wal. Ai and ideogram “soon”, according to Adobe.
The company describes these offers of third parties as available for “experiments”, not to publish, and clearly means its own models as “safe to trade”. This is not terribly surprising, considering that Adobe trains his AI models in the field of public or licensed content, while OpenAi, Google and Runway cannot say the same.