How to Turn Flaude Into Your Personal AI Design Assistant

If you've tried Flaude (or any AI design tool) and the first results felt like generic slop, you're not doing anything wrong. That's the default. AI without rules produces work that's competent but soulless — Helvetica-flavored, mid-gray, no opinion.

The fix isn't a better prompt. It's a better briefing.

Think of Flaude less like a search engine you're querying and more like a junior designer who just joined your team. The junior designer doesn't know your brand. They don't know your fonts. They don't know that your CEO hates rounded corners. They don't know that every CTA in your product uses the same shade of indigo. Until you tell them.

This post is about how to tell Flaude — once — and have it design like you for everything after.

The single biggest unlock: a design system file

Flaude reads from a DESIGN_SYSTEM.md file you create. Every conversation with Claude can reference it. Once it exists, every output runs through it.

You write it once. Flaude follows it forever.

The designer in this LinkedIn thread put it best after a half-day of building one for his brand: *"It's mostly about distilling a designer's design rules and converting them into followable guidelines for the computer. The more robust your set of rules, the better the results."*

That's the whole game.

What to actually put in your DESIGN_SYSTEM.md

Here's the mental model: anything you'd correct a junior designer on, write down ONCE so you never have to correct Flaude on it again.

1. Typography (be specific)

Bad rule: *"Use a clean sans-serif font."*

Good rule:

- Headlines: Anton, uppercase, letter-spacing -0.01em, line-height 1.0
- Body: Inter, regular weight, 16px, line-height 1.6, color #1a1a1a
- Code/monospace: SF Mono, 13px
- Never use Roboto, Arial, or system defaults

The specificity matters. AI fills in gaps with whatever's most common online — and what's most common online is generic. The more constraints, the more your design language emerges.

2. Colors (give it the palette, not vibes)

Bad rule: *"Use warm earthy tones."*

Good rule:

Brand palette:
- Primary background: #FAFAF5 (warm off-white)
- Primary foreground: #1A1A1A (near-black)
- Accent: #3B3BFF (electric indigo, use sparingly for CTAs only)
- Muted: #8888A0 (cool gray for secondary text)
- Never use pure black (#000000) or pure white (#FFFFFF)

Pure black and pure white feel sterile. AI defaults to them. Forbidding them in your rules forces a softer, more deliberate aesthetic.

3. Imagery and visual style (describe the world)

Bad rule: *"Use nice product images."*

Good rule:

Image style:
- Photography over illustration when possible
- High-contrast, slightly warm color grading
- Subjects mid-action, not posed
- No stock-photo smiles, no generic "team meeting around laptops" shots
- When illustrations are needed: hand-drawn line style, single-color, no gradients

You're describing the world your brand lives in. Specifics force specificity in output.

4. Spacing and layout

Bad rule: *"Use consistent spacing."*

Good rule:

- Grid: 4px base unit. All spacing in multiples (4, 8, 12, 16, 24, 32, 48, 64).
- Section padding: 64px vertical, 24px horizontal on mobile, 48px on desktop.
- Cards: 16px internal padding, 12px border radius, no drop shadows.
- Buttons: 12px vertical / 24px horizontal padding, 8px border radius.

This is the kind of detail you'd refine over weeks with a real teammate. Front-load it.

5. Tone of voice (for any copy in the designs)

Bad rule: *"Sound professional but friendly."*

Good rule:

Voice:
- Direct, like a smart friend explaining something. Not corporate.
- Short sentences. Cut every adverb.
- "You" not "users". "We" not "the company".
- No exclamation marks. No emojis unless the brand explicitly uses them.
- When in doubt, sound like a thoughtful Substack writer, not a SaaS landing page.

If Flaude generates microcopy that sounds like every other SaaS in 2026, this rule fixes it instantly.

Bad vs Good: a real example

Bad ask:

"Design an onboarding screen for my meditation app."

What you get back: a purple gradient, a generic illustration of someone in lotus pose, "Welcome!" with an exclamation mark, "Get Started" button. Fine. Forgettable.

Good ask (with DESIGN_SYSTEM.md):

"Design the first onboarding screen for my meditation app. Reference our DESIGN_SYSTEM.md for typography, colors, and tone. The screen should introduce the core promise: helping users build a 5-minute morning habit. Include one photographic hero element (per our image rules) and one CTA."

What you get back: something that looks like *your brand*. Because the rules already told it what your brand is.

The iteration mindset

Almost no one nails their DESIGN_SYSTEM.md on the first try. The designer in that LinkedIn post hit version 5 or 6 before it really clicked. That's normal.

The cycle:

Each iteration tightens the system. After 5–6 rounds, you stop getting surprised.

Treat it like training a teammate

This is the mindset shift that changes everything: you are not "using AI." You are training a junior teammate who happens to be infinitely patient and works for free. The DESIGN_SYSTEM.md is their onboarding doc. The conversation with Claude is the briefing for each piece of work. Your taste is the bar.

The reason this matters: a designer with great taste and a weak teammate produces mediocre work. A designer with great taste and a strong, well-briefed teammate produces *more of their own taste, faster*.

Flaude doesn't replace your taste. It scales it.

Where to start today

If you're brand new to this:

You'll feel the difference in version 2. By version 5, Flaude will start designing things you didn't even realize you wanted, but recognize as yours immediately.

That's the real magic. Not AI generating from nothing. AI generating from *you*.

Try Flaude free →