You've probably heard how Figma Make can knock out UX drafts in half a day, right?
But hardly anyone mentions that moving a single button 50px triggered 64 credits on the invoice.
Yeah, the speed is real
Cut to the chase: the draft generation speed isn't hype. Gondi, a designer at Ohouse, reported that after adopting Figma Make, wireframe work dropped from 3–4 days to half a morning.
Figma's own pitch echoes that. Write your screen and flow in plain language, and it builds the structure, navigation, and interactions all at once. Instead of hand-drawing frames one by one, you start with a sentence like "build a login-to-home flow." That's it.
Field reviewers land in the same ballpark. aiflowreview calls Figma AI a first draft engine. It's not about pixel-by-pixel polish; it's about moving the structural decision — "what goes on this screen?" — into existence before you think. You're no longer staring at a blank canvas; you're refining something real.
Figma Make first shipped at Config in May 2025 and rolled to all paid plans by summer. It's barely over a year old.
Then you look at the bill
On paper, flawless. Once you hit the user community, though, the story flips. The Figma forum is flooded with rants about credit burn.
By the math of users on standard paid plans, 3,000 monthly credits barely covers 45 minutes of real work. One report: 64 credits gone for moving a toast 50px down. Another: over 600 credits for a single complex edit. It adds up fast.
| Task | Credits spent |
|---|---|
| Move toast message 50px | 64 |
| Complex single edit | 600+ |
| Enterprise team weekly usage | 30,000+ |
Here's the kicker, though. One user said "70% of my prompts are fixing AI's first-pass mistakes." The initial draft flies out; the work to make it usable burns the rest of the credits.
The output itself isn't production-ready either. A reviewer who tested it found that Make projects don't open as fully editable layers in Figma Design. Vectors and components are locked solid. You're redesigning anyway.
Ohouse's case hits the same nerve. Yeah, wireframes got fast. But the AI output didn't automatically wire into the company's design system components. The next bottleneck—prepping user test data—landed on Claude Code and MCP instead. One phase got faster; the whole workflow didn't.
| You thought it'd be | Here's what actually happens | |
|---|---|---|
| Draft creation | One-liner and done | One-liner to start; credits bleed through rounds of tweaks |
| Output quality | Ship-ready design | Editable starting point; redesign assumed |
| Design system | Auto-mapped components | Manual mapping required |
How to not leak credits
That doesn't mean skip Figma Make. Knowing where credits go means you can actually save them.
- Name your layers first
Figma says consistent layer naming boosts AI output quality. Use descriptive names like "Button_Next" or "Screen_Home" upfront. - Preview before you confirm
Select a frame, hit Add interactions in the Prototype tab, and the AI proposes a flow. Don't confirm immediately—Preview the actual behavior first, then Keep it. Saves a ton on rework credits. - Stay in the 10–20 round window
Experienced users cap exploratory rounds at 10–20, then move to Figma Design for deeper tweaks. Each additional round racks up credits exponentially. - Narrow scope to genuinely new flows
Works great for features with clear requirements. Doesn't fit mature products with rigid UI standards. Tighter scope = fewer fix loops. - Schedule design system mapping separately
AI output doesn't auto-link to your component library. Right after the draft lands, block time for component swaps.
Go deeper
AI Interactions official guide Figma's own walkthrough—from layer naming through preview and confirm help.figma.com
Figma AI feature roundup Side-by-side pricing and credit breakdown for Make, Agent, Weave, and Dev Mode banani.co
Hands-on user review Real project built in Figma Make, plus honest notes on its limits medium.com
Credit complaint thread Users documenting specific credit-spend cases and frustrations forum.figma.com
Ohouse designer's AI workflow How they split Figma Make and Claude Code in production bucketplace.com
