Adobe has introduced a series of updates to its Firefly Video Model, which enable accelerated workflows through enhancements such as improved motion fidelity and advanced video controls. Aspect rations can be tailored to match format with no extra editing required.
Generative AI partner models Luma AI Ray 2 and Pika 2.2 will be added to Generate Video on Firefly, allowing users to select the best fit model to meet creative needs across sound, video and image. Topaz’s Image and Video Upscalers and Moonvalley’s Marey will launch soon in Firefly Boards, said the company.
New workflow tools enable control of video composition and style and custom generated sound effects can be layered inside the Firefly web app, allowing users to experiment with AI-powered avatar-led videos, said the company.
Style Presets enable the application of distinct visual styles with a single click, speeding up workflows, while Keyframe Cropping accelerates the cropping process with no need to jump between tools. Composition Reference for Video allows users to upload a reference video and add a description, generating a new video that transfers the original composition to the new generation, maintaining visual flow and enabling content to be repurposed.
Generate Sound Effects (beta) provides sound effects from a single prompt, with voice control enabling timing and intensity of the sound to be managed. Once complete, videos can be exported to Adobe Express or Premiere Pro.
Scripts can be turned into avatar-led videos with Text to avatar (beta). A diverse library of avatars, backgrounds, colours and accents are available with Firefly quickly producing videos for a variety of use cases, said the company.
The upgrade also introduces the Enhance Prompt feature within the Generate Video module on the Firefly web. Language is added to original inputs to allow Firefly to better understand creative intent. Ambiguity is removed and direction sharpened to speed up workflows, reducing guesswork and allowing users to spend more time in creative tasks.
Adobe Firefly models are only trained using assets the company has permission to use, in line with its policies to develop generative AI responsibly. According to Adobe, the company never trains on content users produce in its tools.
