Adobe's Firefly AI Assistant, now in beta, is a conversational agent that drives Photoshop, Illustrator and Lightroom on the user's behalf — type a request, and the chatbot picks the tools, executes the edit, and narrates each step. The Verge's hands-on review, published May 29, 2026, found the assistant capable enough to fool a casual viewer but well short of professional output. Adobe is positioning it as a time-saver for creatives; the test results read more like a novice designer's portfolio.
The interface is a standard chat window with a text box and a file-upload button. The assistant doesn't open the desktop apps — instead, it has access to common capabilities including masking, object detection and generative fill. Asked to make a photo more colorful or swap a background, it produces a result and walks through which Photoshop or Lightroom features it called along the way.
On a test image of a cat by a window, the assistant correctly identified the breed as a Maine Coon, described the original scene in detail, and then explained the editing sequence it would use to clear the clouds and brighten the sky. Users don't see the canvas update in real time, but the bot reports which feature it is invoking for each adjustment. That running commentary is the product's most distinctive trait.
Key facts
- 01Adobe's Firefly AI Assistant is a chatbot-style beta that operates Photoshop, Illustrator and Lightroom tools on the user's behalf.
- 02The assistant explains its editing steps in real time, names the specific tools it uses, and flags when an approach isn't working.
- 03It refused to generate illegal imagery, alter body shape, or place the subject in revealing clothing during testing.
- 04When asked to format a social ad, the bot offered TikTok, Instagram and Facebook presets and noted Instagram Square dimensions of 1080x1080p.
- 05Canva shipped a competing conversational design agent that does not narrate its process and produced weaker results in side-by-side testing.
It is also forthcoming about failure. Asked to split objects in a JPG into separate layers, the assistant said it couldn't do that directly, offered two workaround approaches with pros and cons, and then mid-task acknowledged the first method wasn't producing the intended output.
The chatbot redirected, switched to masks and Adobe's crop-and-resize tooling, and completed the request. That self-correction loop is rare in consumer AI design tools, where chatbots more often plow ahead with a flawed approach and present the result as finished work.
“I notice the gaussian blur approach isn't giving me true transparent cutouts — it's outputting full-image PNGs”— Firefly AI Assistant, Adobe chatbot, as quoted by The Verge
Output quality is the weak link. Hair-color swaps, background relocations and lighting changes look convincing at a glance but show oversaturated colors and poorly blended composites under scrutiny. Generative additions — placing new objects into a photo via a Generative Fill-style flow — were visibly subpar compared with non-generative edits like exposure adjustments. The reviewer's verdict was that the work looks like a novice designer's, not a fake that would fool anyone studying it.
Content guardrails held up in testing. The assistant added cigars and even firearms to images when asked, but refused to depict the subject shooting another person, declined to alter the shape of the user's face or body, and would not place them in revealing clothing. It also resisted outright illegal content. That posture stands in contrast to Grok, which has drawn criticism for permissive image generation.
The bot still leans on filler praise — calling routine requests a "great idea" — but most of its outputs are functional information rather than flattery. When the user asked it to turn a cocktail photo into an illustrated bar ad, the assistant asked which social platforms the design was for, surfaced TikTok, Instagram and Facebook as options, and supplied the standard pixel dimensions for each, including 1080x1080p for an Instagram Square post.
Canva shipped a similar conversational design agent recently. The Verge's comparison found Canva's version produced weaker edits and didn't narrate its process, instead taking instructions and iterating silently until the user approved a result. Adobe's choice to expose tool names and editing terminology gives the Firefly assistant a teaching dimension that Canva's lacks — a user who didn't know what a mask was can learn the concept while the bot uses one.
The skeptical read is straightforward: the output isn't good enough for working designers, and people who want to learn the craft are arguably better served by YouTube tutorials. The assistant's value sits in a narrow band — users with middling skills who need a faster path through busywork edits, and beginners who want the task done while they pick up vocabulary. For senior creatives, the reviewer compared the experience to babysitting a new intern.
Adobe's broader bet is that a conversational layer on top of Creative Cloud tools is worth shipping even before the underlying generative quality matches a human professional. The Firefly assistant earns points for transparency, self-correction and refusal behavior — three things most consumer AI image tools handle poorly — but loses them on raw output. If Adobe can close the quality gap without losing the explainability, the product moves from intern to junior designer. Until then, it's a useful demo of what an AI agent inside a creative suite should feel like, wrapped around results that still need a human to finish.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
