One is the text and image to video generation that Adobe previewed last month, accessible in the Firefly web app at firefly.adobe.com. This enables users to create five-second, 720p-resolution videos from natural-language text prompts. These can contain realistic video footage and 2D or 3D animations. It’s also possible to generate video using still images as a prompt, meaning a photograph or illustration could be used to create b-roll footage.
To provide greater control over the output, there are options for different camera angles, shot size, motion and zoom, for example, while Adobe says it’s working on more ways to direct the AI-generated video.
Waiting list
Adobe said it only trains the video model on stock footage and public domain data that it has rights to use for training its AI models. It won’t use customer data or data scraped from the internet, it said.
