Adobe Firefly in Production: 3 Wins and 3 Limits

Adobe Firefly is in 6.5 billion generated images and counting. Two senior designers share where it genuinely saves time — and where it still needs a human to finish the job.

Mayur Domadiya · June 10, 2026 · 8 min read

Adobe Firefly has been used to create more than 6.5 billion images — a number that reflects both the scale of interest in AI-assisted design and the breadth of tasks these tools now cover in professional workflows. Generative Fill in Photoshop, Text to Vector Graphic in Illustrator, and Generative Recolor have moved from experimental features to tools that senior designers use on client work, daily. But the gap between "AI was involved" and "AI handled it cleanly" is still significant in specific areas, and knowing exactly where that gap sits saves the hours you expected to gain. Two senior designers who use these tools on production work for clients including Google, Spotify, Fiat, and AB InBev share what actually ships — and where they still finish the job by hand.

Win 1 — Photo Editing: From 3 Hours to Under 1

Jessica Souza's clearest example is a client brief that most designers will recognize immediately: a highly specific image request. The client needed a woman holding a prescription package in one hand and a phone in the other — happy, standing in front of a pharmacy, wearing casual clothing with green accents. Before Generative Fill, executing that brief meant extensive stock research, purchasing a selection of images, and spending three to four hours compositing them together — and the result still never quite matched the spec.

With Generative Fill, the workflow is: purchase one base image that approximates the brief, then use Generative Fill to modify specific elements — adding the pharmacy background, adjusting clothing color, inserting the prescription package — through text prompts applied to selected regions. Total time from initial search to final file: under one hour.

What changed is not just speed but feasibility. A brief that previously consumed a junior designer's afternoon can now be handled as a task within a larger project day. That productivity shift compounds across a week of client work. For SaaS teams producing ongoing marketing or product imagery at volume, the same arithmetic applies: the economics that previously made custom photoshoots exclusive to well-funded brands now tip in favor of AI-augmented editing for a significantly wider range of teams. Smaller clients can access a level of custom visual quality that was previously out of reach on budget.

Win 2 — Custom Icons and Vectors Without the Stock Library

The conventional icon workflow involves three steps that are not particularly creative: search a library (Google Material, Apple, FontAwesome, or a licensed set), find the closest match, and then edit it into something specific enough to actually serve the design context. Depending on the license, this also carries recurring cost and the risk of an identical icon appearing across competitor products.

Sérgio Estrella's switch to Text to Vector Graphic in Adobe Illustrator eliminates the first two steps for most icon work. Typing a description like "medication bottle, single color, minimal linework" generates a starting point that is specific to the context rather than a generic library asset. The result is a vector that the designer then edits, but the starting point is already closer to the target than anything a stock search would surface — and it will not appear identically on another product.

The practical consequence for product teams is twofold: icon work is faster, and the resulting icon set is more visually coherent with the product's specific context. For early-stage SaaS teams making initial visual identity decisions, this is a cost-effective path to consistency that does not require a dedicated icon designer or a per-icon licensing arrangement.

Win 3 — Client-Directed Color Exploration in Seconds

Client feedback on brand color is rarely precise. They say things like "something more energetic, with the warmth of autumn but not too orange." Translating that into a design direction used to require a designer's full interpretive pass — producing a set of options, presenting them, and then repeating the cycle if the interpretation missed. Each iteration added half a day or more to the review loop.

Sérgio's approach with Generative Recolor is to feed the client's own language directly into the prompt. In one branding project, he took the actual text from a client message, added design-specific descriptors, and submitted it to Generative Recolor in Illustrator. The tool produced several color variation alternatives. The client chose one of them — resolving a review cycle in minutes rather than across a day.

It made me feel afraid for a moment — like AI was going to replace me. But I see this technology as a tool for making adjustments. It didn't replace all my work; it just helped me adjust one aspect to get closer to what the client envisioned.

The mechanism is the same principle behind effective AI UI design: clients do not need to be able to name what they want, they need to recognize it when they see it. Generative Recolor generates the range of options; client recognition closes the loop. The human judgment involved in knowing which option is right — and why — stays entirely with the designer and the client.

Limit 1 — Image Quality Still Trails Midjourney

Both designers are direct about this: for high-end image generation from a text prompt, Adobe Firefly does not yet match Midjourney. The quality gap shows up most visibly in photorealism and compositional coherence — the difference between "almost right" and "right" that is visible to anyone looking closely at the output.

The workflow that accounts for this is layered: use Midjourney or another high-quality generator for initial concept images where raw visual quality is the primary requirement, then bring the result into Photoshop for refinement and integration using Generative Fill. "When I use Midjourney to start the concepts and Photoshop to refine them, then I have the perfect flow," Jessica describes. The two tools handle different phases of the same job.

For product teams building AI-assisted image workflows, the implication is clear: Adobe Firefly is the right tool for modification and extension of existing images, not for generation from scratch when the brief demands a specific level of visual polish. Knowing which job the tool is suited for prevents the experience of running a generation workflow and being disappointed by an output that does not match a Midjourney quality benchmark — because you were using the wrong tool for that phase of the work.

Limit 2 — Vector Output Needs Hand-Finishing

Text to Vector Graphic generates a starting point, not a finished asset. The structural failure mode Sérgio describes is precise: the AI does not build objects, it builds components — "it makes little pieces that it tries to join together." The resulting shapes are sometimes soft, the paths not sharp, and the overall output can resemble a bitmap image that has been auto-traced rather than purpose-built vector art.

A designer using Text to Vector Graphic should budget time for path cleanup, shape merging, and edge sharpening before the vector is production-ready. For simple icons used at small, consistent sizes, this overhead is minimal. For logos, detailed illustrations, or anything that will be scaled across a range of formats and sizes, the cleanup work needs to be factored into the workflow estimate — otherwise the time savings from skipping the stock library search disappear in post-processing.

The honest framing is that AI-generated vectors are currently a better starting point than a blank artboard, but a worse starting point than the output of a skilled illustrator. That positioning is still useful for acceleration, as long as the workflow accounts for it.

Limit 3 — Typography and UX Automation Are Underdeveloped

Adobe Firefly's current feature set is strongest for visual asset generation and weakest in two specific areas that matter most to UX designers: typographic expression and workflow automation for repetitive mockup tasks.

On typography: AI-generated text within images and graphics lacks the spacing precision and expressive range of hand-crafted type work. The issue is often subtle — visible on close inspection, not immediately apparent at a glance — which is the worst failure mode for typography because it gets through review cycles and surfaces in finalized assets.

On workflow automation: the AI capabilities that would save the most time in UX work are largely absent. The missing feature Sérgio describes is contextual content generation for mockups — the ability for the tool to detect that a designer has laid out a list of user cards or an avatar grid and automatically fill the placeholders with generated names, avatar images, and message text. These are not creative decisions; they are repetitive, time-consuming setup tasks that add overhead to every mockup without adding design value. Automating them would shift designer time toward the work that actually requires judgment. That automation does not yet exist in Firefly, and it is the most significant gap for UX-focused workflows.

What This Means

Adobe Firefly has crossed from useful-in-demos to useful-in-production — but the wins are specific, and so are the limits. All three genuine workflow wins follow the same pattern: modifying a starting point rather than generating from scratch. Extending a photo, recoloring a vector, generating an icon variant from a text description — these are all tasks where a human decision has already defined the direction and the AI is executing a transformation within it. The tasks where the tool falls short are precisely the tasks that require generating polished output from a vague brief, or the typographic and spatial precision that comes from years of practiced design judgment.

For SaaS teams making resourcing decisions about design capacity, the practical read is this: Adobe Firefly is now worth incorporating as a modifier and accelerator in an existing design workflow. It is not yet a substitute for a senior designer on generative or high-polish work, but it meaningfully changes what a skilled designer can produce per hour. The second-order effect — that it levels the field between well-funded brands and early-stage teams on access to quality visual output — is probably the more significant one. What previously required a photoshoot budget is now accessible to any team willing to learn which tool handles which phase of the workflow, and where the hand-finishing still has to happen. That is the kind of engineering and design judgment that compounds into a real competitive advantage over time.