New Firefly interface: More effort, less usability

I can’t judge the technical side, but I agree with your point. I haven’t noticed any major improvements in the quality of Adobe’s own AI over the past few months.

What bothers me more than the AI’s quality, though, is the increasingly poor usability. It feels like a toddler playing with building blocks, constantly rearranging things at random.

Testing new ideas is important, but there was a time when that happened internally. These days, it feels like paying customers are being used as beta testers.

I used to look forward to major updates. I genuinely enjoyed trying out the new features. It was like opening a new book — a sense of anticipation, followed by surprise and delight.

These days, I put off installing updates because my first thought is more likely to be: “Oh God, what have they changed this time?”

Come to think of it, I find the partner models more interesting anyway. With Nano Banana, I get better results that are closer to what I’m aiming for than with the Firefly model. I mainly use Adobe’s own model directly in Photoshop to remove elements and retouch images.

For the partner models, Adobe primarily provides the interface. That makes it all the more frustrating when that interface becomes harder to use.

My subscription also renewed for another year in September. Photoshop, Illustrator, InDesign and Premiere are still useful tools for my work, although I could list quite a few problems that have developed over the past months and years. But that is another discussion.

Fortunately, I mainly need images and rarely need video. The credit-based approach to video generation makes little sense to me: spending hundreds of credits on an attempt without knowing whether the result will be usable. By the time I get a suitable clip, my standard monthly allowance could already be gone.

My current workaround is to develop the image with ChatGPT first, then use Adobe to adjust specific details. At least with ChatGPT, I can discuss the idea and clarify what I want before generating anything. With Adobe, it often feels like a gamble whether my prompt will produce a useful result.

Given Adobe’s long-standing role in professional graphics software, I find it particularly disappointing how poorly its AI interprets reference images and translates their visual style into usable results for my work.

On top of the inconsistent results comes this interface: a bewildering mix of buttons, controls, unexpected navigation and unclear icons. Settings seem to change, references disappear when switching views, and I keep having to figure out where I am and how to get back.

I cannot know whether any of this is deliberate. But it increasingly feels as though keeping users inside an ecosystem and promoting additional services takes priority over helping them work efficiently.