On Thursday, Adobe unveiled the latest version of its Firefly family of image generation AI models, including a model designed for generating vectors, along with a newly redesigned web app that features all its AI models and some from its competitors. Additionally, a mobile app for Firefly is in development.
The new Firefly Image Model 4, according to Adobe, offers enhancements over previous models in terms of quality, speed, and user control over the outputs’ structure and style, as well as camera angles and zoom. It is capable of producing images at resolutions up to 2K. A modified and more advanced version of this model, named Image Model 4 Ultra, is able to render intricate scenes comprising small structures and substantial detail.
Alexandru Costin, Adobe’s VP of Generative AI, explained that the models were trained using a higher order of computational power to enable them to produce more detailed images. He also noted that the new models enhance text generation within images compared to their predecessors and allow users to direct the model to create pictures in a chosen style using specific images.
Adobe is also expanding availability for its Firefly video model, which launched in limited beta previously, making it accessible to a wider audience. This model enables the generation of video clips based on text or image prompts, allowing for the manipulation of camera angles, specification of start and end frames, generation of atmospheric elements, and customization of motion design components. Video clips can be generated from text at resolutions up to 1080p.
In parallel, the Firefly Vector Model is designed to produce editable vector-based artwork, enabling the creation and iteration of logos, product packaging, icons, scenes, and patterns.
The web app for Firefly offers access to all these models, as well as image and video generation models from OpenAI (GPT image generation), Google (Imagen 3 and Veo 2), and Flux (Flux 1.1 Pro). Users have the flexibility to switch between models, and all generated images will include content credentials for authenticity. Adobe indicated potential future additions of other AI models to this web app.
Furthermore, Adobe is conducting public tests on a new product called Firefly Boards. This canvas for ideation or moodboarding allows users to generate, import, remix, and collaborate on images, similar to AI-based ideaboards such as Visual Electric, Cove, and Kosmik. Firefly Boards is accessible via the Firefly web app.
Adobe plans to integrate these new models into its product lineup soon, though no specific timeline was provided. The company is also offering its Text-to-Image and Avatar APIs generally, with a new Text-to-Video API available in beta, all accessible through the Firefly Services suite of APIs, tools, and services.
In addition, Adobe is testing a web app named Adobe Content Authenticity, which enables users to attach credentials to their work for ownership and attribution purposes via metadata. This app allows users to specify whether AI companies can utilize any images for AI model training.
