I Gave Figma AI and Brilliant.design the Same 3 Prompts on Free Plans. Here's What Actually Happened

I Gave Figma AI and Brilliant.design the Same 3 Prompts on Free Plans. Here’s What Actually Happened

Same three prompts, same free plans. One tool finished in under a minute, the other kept breaking.

Quick note before you dive in: every tool or feature I mention here, I’ll tell you straight up whether it’s free or paid as of today. I’ve lost count of how many times I’ve watched a video, gotten excited about some AI result, tried it myself, and only then found out the person was on a paid plan the whole time. Not doing that to you.

Last week I wrote about Brilliant.design vs Figma, comparing features side by side. That post covered what each tool can do on paper. This one is different. I wrote three heavy prompts, not quick one-liners, and ran the exact same three through both tools’ free plans. No editing the outputs afterward. Whatever came out is what you’re getting here.

I’ve written about AI design tools before, Google Stitch and Figma Make among them. This test is narrower: not a full app builder, just the AI feature inside Brilliant.design itself, compared against Figma’s own AI, which is still labeled Beta as of today, worth knowing going in.

The three prompts covered three different niches on purpose, so this wasn’t three versions of the same generic app.

A web landing page hero for an AI automation platform

Design a modern landing page hero section for an AI agent orchestration platform called “Pulseflow,” built for teams running autonomous AI agents across their workflows. Use a light lavender-blue background (#EEF2FF). Include: a bold headline in dark charcoal, a sub-headline about connecting agents to real tools, two CTA buttons (primary filled, secondary outlined), a floating dashboard UI mockup on the right side showing active agent cards, each with a task name, a status badge (running, queued, done), and a completion progress bar, a top navigation with logo and 5 nav links, and a trust bar below with 5 company logos. Apply 2026 design trends: glassmorphism cards, subtle grain texture overlay, variable font weight headline, generous whitespace, and micro-shadow depth on components.

A mobile app home screen for a cybersecurity tool, dark theme

Design a dark theme mobile cybersecurity app home screen for iOS, built for monitoring personal device and network threats in real time. Dark background (#0A0E14) with electric blue (#3B82F6) and alert red (#EF4444) accents. Include: a status header showing "Protected" with a shield icon and live threat count, a network activity ring showing blocked attempts/scanned files/active connections, a horizontal scroll of recent alert cards with severity tags, a weekly threat trend chart in bar format, a bottom navigation bar with 5 icons (Home, Scan, Network, Alerts, Settings). Apply 2026 design trends: bento grid layout for stats, frosted glass bottom nav, bold numeric typography for metrics, layered depth with subtle glow effects on accent elements, and dynamic island integration at top.

An admin dashboard for a fintech platform tracking stocks and crypto

Design a modern SaaS admin dashboard for a fintech platform tracking crypto and stock portfolios for retail investors. Pure white background (#FFFFFF) with deep teal (#0D9488) as primary accent. Include: a left sidebar navigation with logo, 8 menu items and user profile at bottom, a top header with search bar, notifications, and settings icons, a KPI cards row showing Portfolio Value ($284,500), 24h Change (+3.2%), Active Positions (17), Cash Balance ($12,400) with trend indicators, a large line chart showing portfolio performance over 12 months, a recent transactions data table with 5 rows showing asset, type, amount, price, and status badge, and a right panel showing top-performing assets list. Apply 2026 design trends: clean data density with generous padding, subtle left border accent on active nav item, color-coded status badges, skeleton loader states visible on one card, and Inter or Geist font system.

Figma: All Three, Under a Minute Each

On Figma’s free plan, you get 500 AI credits every month, with a daily limit of 150. Each full design generation uses about 100 credits, so you get a handful of solid tries each day without hitting a wall.

Figma’s AI feature, called Figma Agents, is still officially in Beta. That just means it’s not finished and polished yet, not that it’s unreliable, and nothing about the results here felt unfinished.

Selecting a frame, clicking the AI icon, and pasting in the prompt was the whole process. All three came back finished in under a minute, no retries, no broken layouts.

Web Hero Prompt

• A two-line headline and a working CTA pair, exactly as asked.
• A dashboard mockup on the right showing agent cards with status badges and progress bars.

Mobile App Prompt

• A phone mockup with a status header, a bento grid of stats, a threat activity ring, alert cards, and a bottom nav bar.
• Everything sat inside the actual phone frame instead of floating loose.

Admin Panel Prompt

• A sidebar, KPI cards with real numbers, and a working line chart.
• A transaction table and a side panel of top performers, all in one coherent layout.

Three prompts, three clean results, all inside Figma’s free plan.

Brilliant: Where It Actually Got Complicated

Brilliant doesn’t hand you a bundled AI model. To run a custom prompt, you need either a connected agent (Claude Code, Cursor, Codex) or your own API key through a provider like OpenRouter.

Quick explainer if you haven’t used OpenRouter before: it’s a single platform that gives you access to AI models from lots of different companies, Cohere, Meta, Google, and more, through one account and one API key, instead of signing up separately with each one. Some models on it are free to use, some cost per use. This is how Brilliant lets you plug in a model without needing a paid coding agent subscription like Claude Code or Cursor.

None of them have a real free tier, so I connected OpenRouter instead and tried three different free models.

Web Hero Prompt, on North Mini Code

• Every word landed on its own line, all the way down the page, instead of wrapping naturally across two or three lines.
• A small model from Cohere Labs, built for speed and agentic coding tasks rather than heavy visual reasoning, and it showed.

Same prompt, retried on Nemotron 3 Super 120B A12B: free

• The dashboard card repeated itself three times. Each copy showing slightly different progress bar states.
• The navigation and trusted by overlapped and became hard to read.

Same prompt again, on Nemotron 3 Nano Omni 30B A3B Reasoning: free

• Took 4 minutes 45 seconds to generate, and card labels broke mid-word.
• When I tried getting it to add an image to the layout, it failed.

Three different free models, three different kinds of broken. Not the same mistake happening twice under a different name.

Mobile App Prompt, on North Mini Code

• The section labels I’d written in the prompt like “Status Header,” showed up as giant literal headings inside the design itself.
• The bottom nav repeated itself several times stacked on top of each other, and nothing sat inside an actual phone frame.

I ran this one a second time to check it wasn’t a one-time glitch. Same result.

Admin Panel Prompt, on North Mini Code

• Internal frames “Frame 44,” “Frame 50,” showed up as literal visible text on the canvas instead of being in the main dashboard frame.
• Several data values came back as question marks instead of numbers, and the sidebar navigation was garbled and overlapping.

A pattern showed up across most of the free models I tried:

• They’re built for text and code, not images. The moment a prompt asked for anything visual beyond plain layout, an icon, a chart, an image element, several of them either broke trying or failed outright with an endpoint error.
• All of them were slow. Every generation took noticeably longer than Figma’s, and some stretched well past what felt reasonable for a single design.

One error showed up repeatedly enough across different attempts that it’s worth naming directly:

“Something went wrong while reaching the AI provider. Please try again.”

This error could hit partway through a generation, not just at the start. And when it did, nothing was saved, the design it had built so far was gone too, not left half-finished for you to pick up. It happened often enough that I’d call it a real reliability problem on the free-model route, not something that happened just once.

Along the way I also ran into OpenRouter’s context window filling up mid-session on a separate attempt, which likely made results worse on top of the model’s own limits.

What This Actually Means

Free doesn’t mean the same thing on both tools. On Figma, free gets you a fast, working result across all three prompts, with no issues, even while the feature is still labeled Beta. On Brilliant, free splits into two paths. The built-in preset prompts (covered in my last post) work fine with nothing connected. Custom prompts are different, they need a connected model, and every free model I tried broke on all three prompts, each in its own way.

To be clear about what this does and doesn’t prove: it doesn’t mean Brilliant’s AI is bad. It means using it for anything beyond the presets is rough right now on the free route, at least with the free models OpenRouter offers. Whether a paid agent like Claude Code or Cursor fixes this, I honestly don’t know. I didn’t pay to test that, so I’m not going to guess at an answer I haven’t checked myself.

If you’ve run heavier prompts through Brilliant with a paid agent connected, I’d genuinely like to know how it went. Drop it in the comments, that’s a gap in this post I can’t fill from here.

I’m Usama, a UI/UX designer based in Islamabad. I write about what design tools actually do, not what they claim to do. More of these coming.

I Gave Figma AI and Brilliant.design the Same 3 Prompts on Free Plans. Here's What Actually Happened was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.