As with most applications being enhanced with GenAI, the creative world of AI-powered image and video creation continues to evolve at a tremendously rapid pace. It’s no surprise then to see Adobe unveil a new version of its Firefly image generation models at the recent Adobe MAX event in London. Nor was it much of a shock to hear that the company had built a new Firefly web-based app (with iOS and Android mobile versions coming soon) around the technology.
What did catch some people by surprise, however, was the company’s move to integrate the image creation tools from several other GenAI model makers – including Google and OpenAI – into the new Firefly app.
As Adobe explained, they heard loud and clear from their customers that, while they very much appreciate all the work Adobe has put into the Firefly models’ ability to generate images and videos with no copyright-related concerns, there are several situations in which they want more options.
For example, if a design firm is working with a client, the designers would likely want to use the copyrighted images, logos, etc., from that company in order to create new materials.
Because of the way Adobe’s Firefly models are trained and built – specifically without copyrighted material – that wouldn’t be possible to do with Firefly. Leveraging image creation models from other vendors, however, could potentially allow those image variations with copyrighted material to be created.
What Adobe didn’t mention, though, is that creative people like to work with as many variations as they can. Bottom line is, they want more options. Because each image generation model works a bit differently, the outputs they generate have their own unique characteristics. It only makes sense then that designers were likely using other models anyway to get access to those different “looks.”
But by making the other models an option within the Firefly app, Adobe retains a bit more control over the creative process. This is especially true now, because new hooks in the Firefly app allow designers to take whatever is created from there – regardless of the model used to create it – and seamlessly pass it over to other Adobe creative apps, such as Photoshop, Premiere, Illustrator, and others. Frankly, it’s a pretty clever way to turn what initially seemed like competitors into something akin to partners.
A related but important development that Adobe unveiled is a new capability it’s calling Firefly Boards-previously announced as Project Concept-that allows designers to put together and share a limitless-sized canvas of images-generated from any model or even copied from any location-in order to stimulate ideas.
These “moodboards” in the Firefly app can then be used to iterate even more ideas with AI, based on the images that are placed upon them. It’s an interesting way for Adobe to dip its toes into the pool of image creation with what could be copyrighted materials. However, Adobe also made it clear that any images created within the Firefly app will carry Content Credentials.
These credentials allow companies to ensure whether the images they create were built with the copyright-free (and thus, safe for public use) Firefly models or with partner (or other non-AI source) models. Again, it’s a great way to let companies experiment, but still provide them with the assurance of the commercial safety that Adobe has built its entire Firefly strategy and technology around.
In addition to these integrations with other image and video creation models, Adobe also announced several major enhancements to its own models. Firefly Image Model 4 is specifically focused on creating even more photorealistic output than its predecessors and offers more options for control than ever before. Still images, for example, can now be created with a wider degree of camera angles, zoom levels, and much more.
In order to encourage more experimentation, the company also offered a new option to create images in a lower resolution 540p format, which can be created much more quickly. Once designers find something they like, they can then crank out a full HD 1080p version for final use. The new Image Model 4 Ultra is a for-pay option with even higher resolution that’s ideally optimized for what Adobe describes as “complex scenes with small structures.”
Both the Firefly Video Model and Adobe’s Vector Model are now generally available, with important refinements in the areas of control options and resolution for each. As with other Firefly models, all the outputs these create can come from a simple text prompt and are commercially safe.
One of the more interesting, and yet completely untouted, capabilities of the output that Firefly generates across all its models is a text description of what it has created. As simple as that may sound, it’s actually very illuminating because it shows how the model sees the output.
To put it another way, it gives you a way to reverse engineer the text prompt that created the image (or video). Given how incredibly important the exact wording of a text prompt can be in an image creation tool, this is a great way to better understand how Firefly “thinks.” It’s also an extremely useful training tool to figure out how to write better prompts as you enter them into Firefly, because prompt training is a skill that I’d argue most everyone is still working on improving.
In addition to its direct image and video creation tools, Adobe also unveiled several APIs that are designed to let enterprise application developers tap into the capabilities of their Firefly models. As you would expect, the Text-to-Image and Text-to-Video APIs do what they say to let companies create applications that can quickly build libraries of creative content. The Avatar API is specifically designed to let companies build explainer-style videos with a digital avatar providing the narration.
All told, it was a solid set of announcements that extend the reach of Adobe’s Firefly tools into some interesting new areas. The company continues to clearly base its core value and differentiation in the area of AI-powered image generation on its ability to create commercially safe content that’s verified to be free of any copyright infringement.
For large organizations, this is a valuable capability. At the same time, Adobe also recognized that the creative users in these organizations-Adobe’s core customers-need more flexibility in terms of how they work with AI-generated material. These latest announcements seemingly make progress towards addressing these needs, and they open up some interesting new potential options for the future.
Bob O’Donnell is the founder and chief analyst of TECHnalysis Research, LLC a technology consulting firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on X @bobodtech
