Adobe has launched the latest iteration of its Firefly family of image generation AI models, along with a redesigned web app that houses its AI models and some from competitors. The new Firefly Image Model 4 improves on its predecessors in terms of quality, speed, and control over outputs.
The Firefly Image Model 4 can generate images in resolutions up to 2K and offers enhanced control over structure, style, camera angles, and zoom. A more capable version, Image Model 4 Ultra, can render complex scenes with detailed small structures. According to Alexandru Costin, VP of Generative AI at Adobe, the models were trained with a higher order of compute magnitude to generate more detailed images.
Video: Adobe
The new models also improve text generation in images and allow users to use reference images to generate pictures in a specific style. Adobe is making its Firefly video model, launched in limited beta last year, generally available. This model enables users to generate video clips from text prompts or images, control camera angles, and customize motion design elements, producing video clips at resolutions up to 1080p.
The Firefly Vector Model can create editable vector-based artwork, including logos, product packaging, icons, scenes, and patterns. The Firefly web app provides access to these models, as well as image and video generation models from OpenAI, Google, and Flux. Users can switch between models and have content credentials attached to generated images.
Adobe is also testing Firefly Boards, a canvas for ideation and moodboarding that allows users to generate, remix, and collaborate on images. Additionally, the company is making its Text-to-Image API and Avatar API generally available, with a new Text-to-Video API available in beta. These APIs are part of Adobe’s Firefly Services collection.
Adobe is integrating the new models into its product portfolio, although a timeline has not been provided. The company is also testing Adobe Content Authenticity, a web app that lets users attach credentials to their work to indicate ownership and attribution.