After spending another year deep in Adobe’s AI ecosystem—testing Photoshop AI and it’s Generative Fill, playing around with Firefly’s standalone tools, and watching these technologies evolve—I’ve got thoughts. Real, practical, this-is-what-actually-happens-when-you’re-on-deadline thoughts.
When we first wrote about the good, bad, and ugly of Photoshop AI, I was cautiously optimistic. Fast forward to 2026, and I’m still feeling about the same — I use the tools a little more than I did before, but only when they can genuinely improve or speed up my workflow.
Is Photoshop AI Good?
Here’s the honest answer: it depends on what you’re trying to do with it.
For extending images? Yeah, it’s actually pretty solid. I use Photoshop’s Generative Fill regularly when I need to expand a photo’s canvas or change a portion of the image. It saves me from having to manually clone stamp my way through backgrounds, and most of the time, it blends well enough that I can move on with my life.
But here’s where it gets tricky. If you’re trying to use Photoshop AI for actual creative ideation — generating a concept from scratch and then refining it — you’re going to hit some significant limitations when using Generative AI in Photoshop.
While writing this article, I was also working on a social media graphic to accompany it. I had a specific visual in mind but wasn’t sure how to prompt for it, so I used Pinterest to identify the aesthetic characteristics I was after — the mood, the lighting style, the motion quality — before building my prompt. I landed on something like: “panning shot of a blurry female silhouette, motion blur, soft focus, film grain, lavender and blue gradient background, stylize 1000.”
The result actually worked well. After looking at it, I decided I wanted it to illustrate the idea of speeding up workflow and working faster on the go, so I added a laptop into the prompt, and it fit well. The key reason the laptop integrated cleanly is that the motion blur and soft focus naturally mask the hard edges where a generated element might otherwise look mismatched. The stylistic characteristics of the image did the heavy lifting. But that’s also the important caveat — this approach worked here because the image style was forgiving. In a sharper, more realistic image, adding an element via Generative Fill in Photoshop AI would be a lot more likely to look off.
This is where the difference between Generative Fill and Firefly’s standalone web app actually matters. When you use Generative Fill in Photoshop AI to edit part of a generated image, it doesn’t have awareness of the full image — it just sees a selection and a prompt, and generates from there. The surrounding context doesn’t carry through the way you’d expect. In my experience with the silhouette image, adding the laptop via Firefly’s web app worked because I was using its “Prompt to Edit” feature, which lets you add a new line to your original prompt and build on the existing image rather than starting from scratch. It understands the scene: the lighting, the motion blur, the atmosphere. That’s why the laptop fits. Generative Fill in Photoshop doesn’t work the same way — it regenerates the selected area without that same full-image understanding, which is why results can feel disconnected or random.
Why Photoshop AI Still Frustrates Designers
The Regeneration Loop Is Real. This is the primary challenge. AI tools promise speed, but when you have to regenerate an image five, six, seven times to get something usable, you’re not actually saving time. You’re just replacing one process with another that requires constant oversight. For a designer managing multiple projects simultaneously, this isn’t just inconvenient — it disrupts your entire flow.
The Prompt Thing Is More Nuanced Than You’d Think. There’s a lot of advice out there telling you to write longer, more detailed prompts to get better results. And sometimes that’s true. But honestly? Some of my best results with Generative Fill have come from leaving the prompt completely blank.
Adobe actually designed it this way. When you select an area and hit Generate without typing anything, Photoshop uses the surrounding pixels as context and fills based purely on what’s already in the image. For smaller edits — removing an unwanted object, filling in a gap in a background, cleaning up an edge — this approach consistently outperforms anything I type in the prompt box.
Where prompts do matter is when you’re trying to add something new that isn’t already implied by the image. In those cases, you need to give it direction, and the results vary a lot depending on how well your vision translates into text. Design is inherently visual — things like mood, balance, and composition don’t always survive the translation into words.
Adobe Firefly and Generative Fill: Same Engine, Different Entry Points
Here’s something worth clarifying, because I see a lot of confusion about this online: Firefly and Photoshop’s Generative Fill aren’t two separate competing tools. Firefly is Adobe’s underlying AI model — it’s the engine. Generative Fill is just one of the interfaces built on top of it, integrated directly into Photoshop. The standalone Firefly web app is another interface. Same technology, different contexts for using it.
The web app makes more sense early in a project — when you’re in the messy ideation phase and want to quickly generate concepts without even opening Photoshop. It’s cloud-processed, so it’s not hammering your local machine. Generative Fill makes more sense once you’re already working inside an existing file — extending a background, removing an object, cleaning up a shot.
But the important thing to understand is that the core limitation — the regeneration loop, the lack of iterative memory — applies to both, because they’re running on the same model. In 2026, Adobe also introduced support for third-party models, including OpenAI’s GPT Image and Black Forest Labs’ FLUX, giving you more style options. But more model choices still don’t fix the fundamental issue of needing to regenerate from scratch every time you want to refine something.
The Bottom Line: Where AI Helps and Where It Doesn’t
For specific technical tasks — extending images, cleaning up photos, and accelerating certain types of edits — these tools provide genuine value. But they can’t understand a client’s brand, make strategic design decisions, or create something that feels unique and authentic.
Real design is about hierarchy, contrast, balance, and storytelling. It’s about knowing when to break the rules and when to follow them, and having the experience to recognize what’s working and what isn’t. At the end of the day, these AI features work best as assistants, not replacements. They handle specific tasks efficiently, but they lack the strategic thinking, creative problem-solving, and human intuition that effective design requires. Use them for what they’re good at, and stay hands-on for everything else.