Adobe Reimagines Generative Video with Latest Firefly

One of the most impressive applications of Generative AI (GenAI) is the ability to create videos from nothing more than a simple text description. Simply type a few words into one of the many different video generation tools now available—including Adobe’s new Firefly Video model—and out pops what can be amazingly lifelike clips. It’s a great example of how powerful this technology has become—as well as how quickly it’s advancing.

At the same time, many of the generative video tools also highlight questions and challenges around content ownership and copyright. Some of these have become a key part of larger discussions on the development and evolution of AI-powered tools. We’ve already started to see this play out with text generation tools that were trained on material scraped from across the internet—much of it being original commercial content. There are serious questions about whether and how content creators should be compensated for their work when it’s incorporated into a large language model. Given the enormous amount of effort (and cost) that goes into creating videos, the voices of concern are bound to get even louder as generative video usage becomes more widespread.

In the graphics world, Adobe recognized these issues early on and made copyright protection a core part of their initial Firefly AI-powered image generation tools. The company chose to only use content which they had licensed and offered compensation to its creators when they integrated it into their model training process. Of course, it helped that they had an enormous trove of content and direct connections to its creators via their long-running Adobe Stock business, which offers millions of still images and videos for sale. Nevertheless, they followed their key principles in leveraging that data and built a set of generative content tools that not only met general rules of fairness and ethics, but also provided a unique guarantee of commercial safety. In other words, anyone who used the Firefly tools were assured that they would not face any legal or financial challenges for using copyrighted content. Given that an important percentage of Adobe’s customers are involved with creating commercial content, that has proven to be an important advantage.

Not surprisingly, as the company makes its latest Firefly Video model available publicly today (though still in beta), it’s following the same commercially safe principles and offering the same guarantees. In addition, as it does with still images, Adobe is integrating support for Content Credentials with its AI-generated video, which means people can reliably verify that it was created with AI—an increasingly critical capability in a world seemingly overrun with deep fakes.

Adobe is offering access to the new model—which was first unveiled last fall—via both a new web application at https://firefly.adobe.com/ as well as through a Generative Extend feature in their Premiere Pro professional video editing program. In conjunction with the launch, the company is also debuting two new commercial Firefly plans and previewing one more. Firefly Standard, priced at $9.99/month, offers 2,000 audio/video credits per month and allows you to create up to 20 five-second 1080p resolution videos per month. Firefly Pro ups the ante to 7,000 credits and up to 70 five-second videos for $29.99/month. Firefly Premium, coming later this year for $199.99/month, is designed for creative professionals that, according to Adobe “expect to generate new video content on a daily basis.”

Like other competitive offerings, the Firefly Video model supports both text and image to video, key frames at the beginning and ending of the clip, and the ability to translate and accurately lip sync audio across 20 different languages. Given that the model comes from a company known for the strength of their creative applications, a key differentiator is seamless integration with other apps across the Adobe suite. So, for example, you can easily create workflows that move from a still image in Photoshop or a vector illustration in Illustrator into the Firefly Video model and integrate the output straight into Premiere. Adobe has also added a new Scene-to-Image tool which can be used to create 3D elements you want to have in the video—whether it’s an animated or photorealistic style.

What’s also clear is that Adobe is focusing on the kinds of tools and capabilities that regular users of its products will appreciate. While many people have been experimenting with some of the other generative video tools for fun, Adobe looks to be focused on delivering practical capabilities that make video creation and editing easier. The new Generative Extend feature in Premiere Pro is a great example of this. While it might only be needed to extend an existing scene for a half second or so, that can make a huge difference for professional editors try to match existing music, audio, and video elements together. Similarly, an early preview of the new Firefly web app user interface shows a focus on the kind of key creative choices for things like camera angle and movements that will help the model generate more interesting and more cinematic type outputs.

While it’s fair to say that Adobe is playing a bit of catch up versus the rapidly evolving competition in the field of generative AI video—and adding 4K support, which Adobe says will be coming later this year, is an example—it’s also clear that the company is applying its own unique approach to the challenge. Not only is Adobe focused on its key licensed content principles, it’s also leveraging its existing toolsets to make integrating the AI-created video content into existing workflows as easy as possible. For the kinds of creative professionals who look to the company for creative applications, that’s an important step.

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on LinkedIn at Bob O’Donnell or on Twitter @bobodtech.