AI is getting better at taking direction
AI tools are getting easier to direct, edit, and fit into real design workflows.
October 14, 2026
Something interesting is happening with AI design tools.
We're seeing more control over image layouts, more choices in AI models, and new ways to automate the repetitive parts of creative production.
That's encouraging, especially if you've spent time trying to get an AI tool to make one small change without messing up everything else.
Here's what caught my attention this week.
Figma Make now lets you try Google's latest AI model
Figma added Gemini 4 Argon to Make's model picker, and it's rolling out across all plans.
What's interesting is that you can switch models during a conversation. If one model isn't getting your layout or interaction right, you can try another without starting over.
Google says Argon is particularly good at complex coding and spatial tasks, including interactive animations.
One thing to watch: Figma says different models can use significantly different amounts of AI credits. Start with something small before using it on a larger project.
FLUX 3 lets you design the composition before generating
This one caught my attention because it approaches image generation more like a designer would.
Instead of describing an entire composition and hoping the model gets it right, FLUX 3 Image lets you draw boxes to define where elements should go.
If you've ever tried to generate campaign imagery with very specific placement requirements, you can probably see why this matters.
The company says edits can leave everything else unchanged. That sounds promising, but I'd still check the results carefully before using them in production.
Claude can now create animated explainers
Claude Motion is a new beta feature that creates short animations from a prompt.
What's different is that it doesn't generate video footage. It writes code to animate text, shapes, charts, and images. That means you can change the actual text, numbers, and timing instead of regenerating an entire video. You can export the result as an MP4 and continue working on it in another editor, including Adobe Firefly.
I can see this being useful for presentations, quick product walkthroughs, and internal communications.
It's currently limited to Team and Enterprise plans, so not everyone can try it yet.
Claude Design, Docs, and Slides also came out of beta this week and are now available on the Free plan.
AI is coming for the repetitive parts of ad production
If you've ever had to turn one approved campaign into a dozen different ad sizes, this one will sound familiar.
Luma's Ad Variants can take a finished static ad and adapt it for different placements and languages. Higgsfield is working on something similar, including tools that create static ads from a website and generate variations of existing video ads.
I'm interested in how well these tools preserve the original design since resizing isn't always as simple as moving elements around. Typography, hierarchy, and composition often need individual attention.
If AI can handle a reasonable first pass, though, that could save designers a lot of production time.
Worth trying: let FigJam organize your research notes
FigJam AI can take a board full of sticky notes, group them by topic, and summarize what it finds.
If you've ever worked through a large set of interview notes or usability findings, you know how much time that first round of organizing can take.
The important part is knowing where to stop trusting the AI. It can group similar observations, but that doesn't mean it understands which patterns actually matter.
Try it on a past research project you've already analyzed. Compare what it finds with your own conclusions and see where it helps or misses something.
One caution: research notes can contain sensitive participant information, so check your data-handling requirements before using AI to process them.
The Pulse: an AI-generated logo still needs a designer
A client shows up with a logo they generated using AI and asks you to turn it into a usable vector file.
The problem is that AI-generated logos can look perfectly fine at first glance while hiding all kinds of issues. Uneven curves, inconsistent geometry, strange lettering, and details that won't survive at smaller sizes.
There's also the question of ownership and what the client can legally protect.
Knowing how to evaluate, refine, rebuild, and prepare that design for real-world use is still a valuable skill.
One more thing: Photoshop can now relight a photo
There's a new Relight feature in the Photoshop beta that lets you add and position up to three virtual lights.
You can adjust their color, intensity, and softness, then render the result on a separate layer.
I can see this being useful for product photography and compositing, especially when the lighting doesn't quite match.
The catch? Each final render costs 30 generative credits.
That's worth keeping in mind if you're experimenting with several lighting directions.
That's it for this week.
I'll keep watching what's changing, testing what looks useful, and filtering out the stuff that probably isn't worth your time.
Until next Wednesday.