On Tuesday, Adobe Inc. revealed its plans to launch a new generative AI-powered video creation tool, Adobe Firefly Video Model, in a limited beta release later this year. With the new tool, the company is expanding its suite of Firefly AI-powered applications into video, aimed at making life easier for professionals in design, photography, and video.
The Firefly Video Model, in particular, will enable users to generate video clips based on text or image prompts and is engineered to react to specific directions like camera angle, panning, zoom, and motion. According to an interview with Adobe's Vice President of Generative AI, Alexandru Costin, the model has been fine-tuned to the quality expected by videographers.
“We've invested in making sure we pay attention to the prompts and respect the guidance from professionals much better than other AI video models on the market," said Costin.
That puts the new Adobe tool into the fast-growing market for AI-powered video creation that is already occupied by other technology firms like OpenAI's Sora and Stability AI's Stable Video Diffusion. While these tools do allow users to generate short video clips from AI models, what Adobe is going to try to do is higher quality that meets the specific needs of professionals.
In turn, the Adobe Firefly Video Model generates five-second video clips with one prompt and is engineered to function seamlessly not only with text but also with images. The model is trained on licensed material or public domain content from the great database of Adobe Stock, which includes more than 400 million images, illustrations, and footage. As hinted by Costin himself, the model does not train on any customer data for compliance with all intellectual property laws.
Along with the generative video tool, Adobe will also introduce Generative Extend, made for its Premiere Pro video editing software. Here, Generative Extend can seamlessly extend existing video clips with up to two seconds of additional footage, making gaps disappear where incomplete content might be apparent.
This makes the work of editors quite easy, whereby the extension of clips won't require shooting or even the headache of finding similar stock footage. The move is likely to resonate with professionals who often have to deal with incomplete footage, allowing them to save time and creative flow.
Many have been concerned about the ethics of deploying generative AI across creative industries. For that, Adobe has assured the world that the Firefly Video Model is trained solely on content either in the public domain or licensed for use by Adobe Stock. Costin put weight on how such a strategy would ensure that no infringement of intellectual property rights occurred and that no copyrighted material would pop out unauthorized.
"We've cleaned up our database to make it so that there are no trademarks or easily recognizable characters that would result in some kind of lawsuit," Costin added.
While AI has stormed the bastion of creativity in every other industry, the Firefly Video Model may mark a game-changing moment for Adobe in video content creation.
With its overall speed, precision, and creative flexibility, Adobe's Firefly Video Model has great potential to be a serious resource for videographers, content creators, and marketers alike. The industry will be closely watching the limited beta release later this year as Adobe positions itself for leadership in professional generative AI applications.