Adobe is trying to solve a very familiar AI problem: the model gets you 80 percent of the way there, and then stubbornly refuses to nail the last 20 percent, no matter how many times you rewrite the prompt. With Precision Flow and AI Markup in Firefly, the company is basically admitting that prompt gymnastics aren’t enough anymore — and is giving creators something much closer to real, hands-on control over AI images.

Instead of positioning these features as yet another “smarter model,” Adobe is framing them as new control surfaces for Firefly’s image editor: one that lets you dial in how much an edit should change the image, and another that lets you literally draw where and how those edits should apply. It’s a subtle shift in marketing, but an important one for anyone who has tried to use AI art tools in real workflows and discovered that “make it a bit moodier” can sometimes turn a pleasant sunset into a full cyberpunk apocalypse.

At the heart of this update is Precision Flow, currently in beta inside the Firefly image editor. The idea is simple: you upload or generate an image, describe the change you want — warmer light, autumn instead of summer, more trees in the background, different time of day — and Firefly generates a spectrum of results from subtle to dramatic that you can scan through with a slider. Instead of re-prompting over and over, you get a controlled range of interpretations in one go and simply stop at the version that feels “right.”

Under the hood, Precision Flow is still using Firefly’s generative models, but the experience is more like color grading or tuning a filter than having a conversation with an unpredictable chatbot. You can select the whole image, a specific region, or brush over an area, then use Edit strength to decide how forcefully the change should be applied. Want just a gentle shift from warm afternoon to slightly cooler dusk? Keep the slider low. Want to turn a clear, bright field into a moody night scene with storm clouds and heavy shadows? Push it further along the spectrum and let Firefly generate more radical variations.

For working photographers, designers, social media teams, or anyone shipping visuals on a deadline, that slider-based workflow matters more than it might sound. In many AI tools, changing “a little” and changing “a lot” both look the same at the prompt level — you just type a new sentence and hope the model interprets your intention. Precision Flow makes nuance visible: you see the “too little,” “too much,” and the “sweet spot” side by side, without losing your original image in the process.

The other new feature, AI Markup, is where Firefly starts to feel less like a text-based magic trick and more like a proper image editor augmented with generative AI. Instead of typing a prompt and praying that the model understands which part of the frame you meant, you literally mark it up: draw, select, annotate, and tell Firefly, “change this, not that.”

Inside the Firefly image editor, AI Markup lives as a set of tools: a brush for freehand strokes, a region/rectangle tool for selecting specific zones, and text boxes you can drop exactly where you want the edit to occur. You can sketch in a tent on a camping scene, outline the area where you want the sky to change, or draw over a product while telling Firefly “make this label more readable” – and the model uses both your markings and your text to generate the new version.

Compared to pure prompt-based tools, the advantage is obvious: spatial intent is no longer a guess. If you only want to tweak the light on a subject’s face while leaving the background untouched, you can do that by brushing over the face and writing a short, localized prompt — not by trying to outsmart the model with yet another long, descriptive sentence. If you want an extra tree on the left, not trees everywhere, you draw on the left.

Adobe’s own documentation leans into this idea: AI Markup is meant to combine selection tools that creatives already know with prompts that generative models need, so that Firefly has both the “where” and the “what” for each edit. You can reset all markups and start from scratch if you change your mind, or build up multiple prompt regions for complex, multi-step changes across a single image. For production work, that’s a big deal — it lets teams iterate in a more surgical way, instead of regenerating entire scenes every time a stakeholder asks for a small tweak.

Taken together, Precision Flow and AI Markup are very much about pushing Firefly beyond the “type a prompt, hope for the best” phase. Adobe has been steadily layering more traditional editing concepts into its AI tools — from Generative Fill and Generative Remove to Generative Expand, Generative Upscale, and Remove Background — and these two new features plug right into that broader strategy. Firefly is increasingly positioned as an all‑in‑one creative AI studio that sits alongside Photoshop and other Creative Cloud apps, not a novelty website you only open to generate wild concept art.

This also lines up with Adobe’s push toward more reliable, commercially usable AI. Over the past year and a half, the company has rolled out improvements to Firefly-powered features in Photoshop — higher‑resolution results, better prompt matching, more consistent reference handling — and has pitched Firefly models as “commercially safe,” trained on licensed and Adobe‑owned content. Features like AI Markup quietly reinforce that promise: if brands are going to trust AI in their imagery, they need predictable control over what changes and where, especially around products, faces, and logos.

Of course, none of this magically erases the underlying limitations of generative models, and even Adobe-adjacent creators are quick to point that out. Early commentary and hands‑on coverage highlight that tricky areas like reflections, glass, chrome, hair, and motion blur can still trip up Firefly, even with localized markups. AI may now better understand that you only want the background changed, not the product in the foreground, but getting pixel‑perfect realism in challenging textures still requires human judgement — and often a pass through traditional tools like Photoshop’s manual retouching tools.

Where these features shine most is in that middle space between pure ideation and final polish. If you’re mocking up a campaign direction, exploring looks for a thumbnail, iterating on a mood board, or pitching a client on multiple visual directions, Precision Flow and AI Markup can compress what used to be long back‑and‑forth cycles into a few minutes of guided experimentation. You can rapidly explore “more drama,” “less clutter,” “cooler palette,” or “different season,” then lock in one version and refine specific regions without restarting the whole process.

All of this is wrapped inside the broader Firefly platform, which Adobe is continuing to expand with new image and video models, better realism, and more ways to go from still images to motion. For now, though, this particular update feels less about raw model power and more about editing ergonomics: giving creators the kind of knobs, sliders, and masks they’ve relied on for years, but wired into a generative engine that can invent pixels on the fly.

If you’ve bounced off AI image tools because they felt too random, Firefly’s Precision Flow and AI Markup are very much aimed at you. They won’t completely replace a skilled editor in Photoshop, and they don’t magically fix every AI artifact, but they do make the whole process feel less like a game of prompt roulette and more like actual image editing — just with a lot more power under the hood.