Adobe unveils Firefly 2, new AI features: what you need to know


VentureBeat presents: AI Unleashed – An exclusive executive event for enterprise data leaders. Network and learn with industry peers. Learn More

Creative software giant Adobe has been walking a bit of a tightrope when it comes to generative AI: while it has embraced the technology to release a host of new features for its users, such as the well-received Generative Fill in Photoshop and Firefly text-to-image generator (both unveiled in spring 2023), it has faced criticism from some contributors to its Adobe Stock image service, who say the company took advantage of its permissive terms-of-service to allow it to train its proprietary AI models on their work without advance knowledge or direct compensation.

But none of that has slowed Adobe down in its embrace of GenAI. In fact, this week at its annual Adobe MAX conference in Los Angeles, Adobe has announced a host of new AI products, services, and features, including its all-new “Firefly Image 2,” which includes improved prompt understanding and greater photorealism, putting it right up against other leading generative AI models such as Midjourney and the recently released DALL-E 3 from OpenAI, now integrated into ChatGPT Plus.

Adobe Firefly 2 screenshot
Adobe Firefly 2 screenshot. Credit: Adobe

Firefly Image 2 offers more enterprise-friendly features, intensifying competition with Canva

While Firefly 2 doesn’t offer baked in-typography like DALL-E 3 or rival Ideogram does, it does offer some other unique and enterprise-grade features.

Among them are Generative Match, a new feature that is similar to the “style transfer” AI art filters that were popular on social years ago, but more sophisticated and advanced, allowing the user to generate imagery in a particular style from a reference image they supply.


AI Unleashed

An exclusive invite-only evening of insights and networking, designed for senior enterprise executives overseeing data stacks and strategies.


Learn More

As Adobe states in its news release: “Generative Match enables users to either pick images from a pre-selected list or upload their own reference image to guide the style when generating new images…Users can easily meet brand guidelines or save time designing from scratch by replicating the style of an existing image, and quickly maintain a consistent look across assets.”

This feature intensifies the already growing rivalry between Adobe, the longtime leader of creative software for visual artists and designers since the dawn of the PC age, and Canva, which over the last decade since launch has won a huge following courting the “everyone else” of the equation, those without art degrees who nonetheless need to create visual material, such as marketers and communications professionals.

Just days before Adobe MAX, Canva pre-empted its competitor’s conference (smartly) by announcing its own new Magic Studio complete with a number of AI features, including a similar feature called “Magic Morph” as well as a writing-driven one called “Magic Morph,” and a text-to-video GenAI feature in partnership with startup Runway.

This week at MAX, Adobe hit back, not only with Firefly Image 2, but with a “New Firefly Design Model” that enables users — and here Adobe specifically calls out “SMBs and enterprises” in its news release language — to instantly generate “stunning design templates,” that they can use in print, social media, and online advertising. As Adobe states:

“Top global brands are also working with Adobe to explore how Firefly drives productivity, reduces costs and accelerates their content supply chains. Adobe and NVIDIA recently announced plans to make Firefly available as APIs in NVIDIA Omniverse, a platform for Universal Scene Description (OpenUSD) workflows, enabling developers and designers to simplify and accelerate their workflows.”

Firefly Image 2 offers something Canva does not yet: “Content Credentials,” a labeling mechanism through Adobe Creative Cloud that applies metadata to imagery signifying its source and, in this case, the fact that it was AI generated or based on a reference image.

And, in perhaps the most enterprise-friendly feature of the event, Adobe introduced GenStudio, a new generative AI powered program allowing companies to customize and fine-tune Firefly for their needs, and control how their employees use it and which Adobe programs have access to it using APIs.

“For example, with 10 to 20 images, teams can instantly tailor Adobe’s powerful Firefly models and enable anyone in an organization to generate on-brand content that is designed to be safe for commercial use,” Adobe writes in its news release on GenStudio. “Additionally, strict governance and security controls ensure a brand’s content, data and workflows stay within the organization.”

Firefly 2 comparisons and examples

Already in the day since Firefly Image 2 was announced, the AI art community on X (formerly Twitter) has been alight with examples of how the new generative AI image creation service compares to its competitors, Open AI’s DALL-E 3 and Midjourney.

AI creatives and influencers such as Chase Lean, Alie Jules, Rowan Cheung, Paul Convert, and Nick St. Pierre aka “nickfloats”, have all posted threads or images comparing Firefly 2 to other leading AI image generators.

While some believe Midjourney remains better at generating photorealistic imagery, others counter that Firefly Image 2 has surpassed it, making photorealistic images look less staged and more candid/authentic.

Regardless, all seem to agree it is a big step up from the first version of Firefly.

Rights and indemnification

Notably, in order to discourage misuse or use of another artist’s copyrighted imagery for unauthorized use by an individual, brand, or enterprise, Adobe says “Generative Match automatically prompts users with an in-app message, requiring them to confirm they have rights to use uploaded images as well as agree to Adobe’s Terms of Use.”

While this confirmation may not prevent bad actors from simply lying and proceeding to use the tool, it shows Adobe’s attempt to be complaint with copyright, an ongoing contentious issue in the age of generative AI, with numerous creatives suing other companies such as OpenAI and Meta Platforms (parent of Facebook, Instagram, WhatsApp, and Oculus VR, among others) for training on their material without express permission, authorization, or compensation.

In fact, Adobe’s new product announcements at MAX this week are replete with many mentions of and gestures toward respecting copyright, including this: “Firefly Image 2 is trained on licensed content, such as Adobe Stock, and public domain content where the copyright has expired.”

Don’t forget, Adobe is one of the short list of companies offering AI tools along with indemnification — that is, compensating or defending users who face claims for the use of said AI tools. Guess who else rolled out generative AI indemnification this month? That’s right, Canva through its new Canva Shield.

A host of additional AI features, including Project Stardust

While Firefly Image 2 was clearly the stare of Adobe MAX so far, the company didn’t stop there. It previewed “Project Stardust,” an object-aware photo editing feature in which generative AI powers new capabilities like dragging objects around an image, changing their colors and positions, and having AI algorithms automatically manipulate the resulting image to make it look intentional and professional.

For those familiar with existing Adobe programs like Photoshop, the company analogizes that the new feature treats “any image like a file with layers.”

This too, is in competition with Canva’s Magic Studio, and specifically its “Magic Grab” feature, which offers similar functionality.

In addition, Adobe unveiled a “Text to Vector Graphic” image generator in its existing Adobe Illustrator drawing program, allowing users to simply type a text prompt in, and Adobe’s AI will nearly instantly generate a range of vector images to choose from. This is useful for designing assets for display on digital devices, as vector images are based on geometric shapes and can usually be resized without quality loss. It’s already receiving praise on X:

Adobe also previewed Project Fast Fill, which uses Adobe’s Generative Fill from Photoshop but inputs it into motion graphics in Premiere Pro and After Effects; Project Dub Dub Dub, which allows audio or video with audio to be automatically dubbed into a number of languages while preserving the speaker’s voice, going up against startups such as ElevenLabs and Captions; Project Scene Change, letting video editors combine two videos from different source cameras into a single seamless transition, automatically connecting them through algorithmically generated camera motions; Project Res Up to convert low resolution video to high res; Project Poseable, a text-to-3D object AI generator that can generate 3D figure poses from 2D imagery; Project Neo for adding 3D to 2D designs; Project Primrose, which allows real clothes to display computer graphics; Project Glyph Ease for AI generated “glyphs” or letters that can be further edited in Illustrator; and Project Draw & Delight, a tool that uses AI to turn real life doodles into “polished and refined” digital sketches.

The preview segment of the conference, called Adobe Sneaks, was hosted by Pitch Perfect and Workaholics actor Adam Devine, and can be streamed online free here.

While these features are only in preview for now, there’s a good chance many will make their way into active Adobe software.

Of course, as with anything new — especially involving generative AI and Adobe — there were vocal critics of the announcements as well.

Criticism and controversy aside, Adobe is clearly committed to AI and sees it as a major part of nearly all its Creative Cloud (formerly Creative Suite) offerings going forward.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.


Source link