Adobe launched a video creation artificial intelligence (AI) model, heralding full-scale competition with OpenAI, Meta, and Runway.
The Verge reported on the 14th (local time) that Adobe runs on the ‘Firefly Video’ model: ▲Generative Extend ▲Text to Video ▲Image to Video It was reported that it had been released.
This release was already announced last September. Amongst them, the official release is a creation extension tool that may extend input video by as much as 2 seconds within the paid version of Premiere Pro.
This feature creates additional moments within the scene, seamlessly linking camera movement and subject movement. AI takes care of any missing parts in the whole video shoot and naturally connects the back and front. Because of this, it’s attracting attention because there isn’t a must take separate shots for the few seconds where there’s a spot.
Prolonged clips will be created in 720p or 1080p at 24 frames per second. Background audio can also be prolonged, marking the debut of the AI audio model being developed by Adobe. Nevertheless, to avoid copyright issues, no voice or music is created.
On the Firefly website, you possibly can beta test the ‘Text-Video’ tool, which creates a video by entering text, and the ‘Image-Video’ tool, which creates a video based on a picture. Each tools can produce AI-generated videos at 720p at 24 frames per second, as much as 5 seconds long. The online beta version is out there free of charge, but usage is predicted to be limited.
Adobe said it has trained Firefly to create various types of media, from animated content to realistic videos. What’s noteworthy is that it is feasible to supply videos containing text, which was difficult for AI image generators. Setting options are also provided to regulate camera movement, angle, shot size, etc.
Creation extensions, text-to-video, and image-to-video all take about 90 seconds to create, and Adobe said it’s developing a ‘turbo mode’ to shorten this.
Since Firefly was learned using open licenses, public domain, and Adobe stock content, the most important advantage is that it’s commercially protected. This will reduce concerns about copyright infringement.
Moreover, an ‘AI generated’ watermark is robotically inserted into the metadata of videos created with Firefly. That is to permit platforms or individuals to find out the authenticity of content using AI identification tools. Images and videos containing drugs, violence, politicians, and copyrighted material were excluded from the educational process.
In this manner, Adobe is getting off the bottom faster than originally expected, and full-fledged competition is predicted to unfold within the video creation AI field. Since news broke that OpenAI is relearning Sora, Google, Meta, Runway, Pica, etc. have been releasing upgraded versions one after one other.
Reporter Park Chan cpark@aitimes.com
