Adobe has spent the past couple of years trying to convince us that generative AI belongs in the creative process, and at Adobe Max in Miami this week, it’s going all in. Here’s the thing: Firefly is no longer a text-to-image toybox for mum’s designing birthday invites. It’s being rebuilt as a proper, all-encompassing AI creative studio that pulls Adobe’s own tools, a stack of third-party AI models and a new Creative Agent into one place, and Adobe’s pitch is simple: if you bring the vision and judgement, its AI sidekick, the agent, will do the legwork.
Rather than prompting a model for one AI-made image, you describe what you want in plain language, and the new Agent orchestrates Adobe’s tools and a growing list of AI models to get there. You stay in the director’s chair, with the option to review the agent’s plan before it starts, pick a model yourself or leave it on auto-select, and step in whenever you like. Adobe believes this workflow will bring new projects within reach that were once out of reach and builds on Firefly and other Adobe AI tools added to Photoshop, Illustrator and Premiere Pro over the last three years.
It's a similar Agentic approach we learned about when Nvidia announced its plans for NVIDIA RTX Spark, to connect 3D apps in one AI-controlled space.
Firefly is now design everything
Firefly Design, in beta, is a new editor built from scratch, with vector, image editing, animation, 3D and print on one collaborative canvas. You can work with AI, use Adobe’s precision tools, or move between the two. The image editor gets new tools for objects, text and lighting, including Edit Object, Edit Text, Relight and Retouch, plus a gallery of dozens of Quick Actions for small jobs like removing an object or knocking out a thumbnail.
In a demo ahead of Adobe Max, I learned how it all links up. A Firefly Boards project, with fonts from Adobe Fonts’ library of around 35,000 typefaces, carried straight into the Firefly Video Editor. Adobe Stock supplied the music, ElevenLabs the text-to-speech, and the new cut detection did in seconds what Adobe’s own presenter says can take minutes, even hours. Title motion presets are in there too.
Firefly also offers models from Google, Kling, OpenAI, ByteDance and ElevenLabs alongside Adobe’s own and Topaz Labs’, with recent additions including Kling 4.0 and Dreamina Seedance 2.5. If you already pay for several of these services, seeing them under one roof will appeal. But this also raises more questions about Adobe’s famously ethical AI approach, as the foxes are now in the henhouse. I’ll be interested to see how this plays out.
Back to the details, Firefly Graph is in beta and lets you string tools and models together, save the result and run it again whenever you need to. Builder is still in private beta; you type a prompt, and it spits out a clickable prototype for a site or app, and Adobe claims it won’t wander off your design system. That’s the claim I’d test first, because most AI prototyping tools drift the moment you ask for something unusual.
Adobe is also previewing ‘always-on agents’ (sounds ominous) that keep recurring work moving while you’re away, like organising assets and checking quality. For example, a restaurant owner gives an agent their menus, URL, and brand assets. The agent drafts weekly social posts for approval, learns from feedback, and reports back on performance.
Creative Cloud gets the agent too
The agent is also spreading beyond Firefly into Adobe’s flagship desktop apps, including Illustrator, Premiere, InDesign, After Effects, and Frame.io, though the AI Assistants there are still in beta (in Photoshop only, in the Photoshop beta app). What caught my eye is that Adobe is also opening the door to rivals. The local MCP integrations, previewed now and due in private beta in early 2027, will let you point your preferred third-party agent at native Photoshop, Illustrator, Premiere and After Effects project files on your own computer. Layers, vectors, bins and sequences stay editable, so you can step in and fix things yourself. Plenty of pros already use outside agents, and Adobe clearly wants its apps to do their work where they do it.
Away from the agent, the app updates keep coming. After Effects gets an AI assistant that can make sense of an inherited project, debug expressions and generate images and video. Photoshop gets Relight, Generative Restyle and Quick Compare, and its remove tool now spots associated elements such as shadows. Lightroom adds Glow, one-tap blemish removal and a beta Open Eyes fix for dodgy blinks. Separately, Topaz Labs, which Adobe recently acquired, has introduced two new models, Wonder 4 and Image HDR, for upscaling, denoising, deblurring, sharpening and HDR. Illustrator gets refreshed content libraries and new on-canvas Blend controls, and InDesign gets a co-editing beta, Flex Layout for text frames and rewrite tools.
What's the Firefly Design catch?
The idea that the artist stays in charge is Adobe's way of easing bad news for any creative fearing how AI is affecting them, and I can believe that for filing, resizing, and checking. Whether that holds once the agent makes calls on composition, layout, and style is the big question, and the more capable these tools get, the blurrier that line will be.
Adobe still has to show that more automation really does mean more time spent making things, and that comes down to how much control creatives get once the demos are over. More importantly, with agentic AI now a trend and spreading to more and more creative software, artists and creatives who want to stay AI-free will find it harder than before.
