What to Type for AI

How to Write AI Prompts for Design That Work First Try

You send a prompt to Midjourney. It comes back wrong. So you rewrite it. And rewrite it again. Four hours later, you've burned through revisions and still don't have what the client asked for.

The problem isn't the AI tool. It's that most designers write prompts the same way clients brief them—vaguely. "Modern and clean." "Bold but professional." "Something fresh."

AI needs specifics: exact color ranges, material textures, spatial relationships, lighting direction, and style references. When you include these details in a structured order, you get usable outputs immediately. This page shows you the exact framework that cuts revision loops from 15+ hours per client down to one or two rounds.

Why Your AI Prompts Aren't Working

A vague brief produces a vague prompt. If a client says "make the packaging look premium," you probably write: "luxury skincare packaging, elegant, gold accents." The AI interprets "luxury" differently every time. One render looks like drugstore gloss; the next is funeral-home baroque. The fix: decode what "premium" actually means for *this* project. Is the customer wealthy and minimal, or wealthy and maximalist? What colors signal that taste? What materials? What level of detail? These specifics go into your prompt in a consistent order, and suddenly the AI understands the assignment.

The 5-Step Brief Decoding Process

Before you write a single prompt, extract these five things from the client brief: **1. Core product or service.** What is it, exactly? "Skincare" is too broad. "Retinol serum for sensitive skin" is workable. **2. Target customer (in one sentence).** Who buys this? "Women 25–35 who read wellness blogs and spend $60+ on skincare." **3. Competitor tone.** Name one brand they like and one they don't. "We want Augustinus Bader's minimalism, not Tatcha's ornate feel." **4. Specific visual constraints.** Color palette range (not just "gold"—try "warm champagne to deep bronze"). Materials (matte, gloss, texture). Typography weight. Negative space ratio. **5. Intended use and output size.** A 3D mockup lives on Instagram at 1080px. A packaging template needs 300dpi for print. Context changes what details matter.

The Master Prompt Formula (In Order)

Once you've decoded the brief, structure your prompt like this: **[PRODUCT] for [CUSTOMER TYPE]** [Specific style reference or mood, not generic adjectives] [Color palette with hex or natural ranges] [Materials and finishes] [Composition and spatial layout] [Lighting direction and intensity] [Output format and dimensions] Example (generic vs. structured): *Generic:* "Modern skincare packaging, clean, luxurious, gold accents." *Structured:* "Minimalist retinol serum bottle for dermatology-conscious women 30–45. Visual reference: Augustinus Bader's restraint, not Tatcha. Matte white glass with brushed stainless steel cap. Warm champagne to deep bronze accents on label only—no packaging pattern. Centered sans-serif typography, ample negative space. Soft top-left directional light, no shadow. 3D product mockup, flat lay, white background, 1080×1200px." The second one is longer but it cuts revision rounds in half because the AI has nowhere to guess.

Common Prompt Failures and One-Line Fixes

Even with the formula, certain phrases break AI outputs. Here's what to watch for: **"Beautiful" or "stunning" or "professional."** These mean nothing. Replace with specific references: "in the style of [brand or artist] but with [one constraint]." **Multiple competing styles in one prompt.** "Modern and vintage" confuses the tool. Pick one dominant style and say what you're borrowing from the other: "Modern primary style; vintage film grain texture, 10% intensity." **Vague lighting.** "Well-lit" produces flat, generic images. Say where light comes from and how hard it is: "Hard side light from 45° left, deep shadows, 3:1 contrast ratio." **No negative constraints.** Tell the AI what *not* to do. "No patterns, no drop shadow, no serif fonts, no warm tones." **Output format buried.** Put dimensions and file type at the end: "3D product render, glossy finish, 2400×3200px, 300dpi, white background." This prevents the AI from returning a flat illustration when you need a mockup.

When to Use Midjourney vs. DALL-E

Both tools follow the same prompt structure, but they excel at different things. **Use Midjourney for:** product packaging, 3D mockups, complex lighting scenarios, artistic style consistency across variants, and anything where you need precise 3D form. Midjourney's rendering is stronger and more controllable. **Use DALL-E for:** flat graphic design, illustration, marketing collateral, composited scenes with multiple objects, and rapid iteration on color/composition without re-rendering geometry. For a skincare bottle with label, Midjourney gives you a render you can drop into mockups. For an Instagram ad layout with typography, DALL-E is faster and easier to iterate. The Master Formula works for both—just add tool-specific modifiers. For Midjourney: "--ar 3:4 --v 6" for aspect ratio and model version. For DALL-E: "in the style of [illustrator or art movement]" usually works better than studio lighting descriptions.

Building Prompt Variants From One Approved Brief

Once you get a first-draft output the client approves, you can generate variations without starting over. Keep your Master Formula prompt and change only one or two elements: **Color reskin:** Swap out the color range. "Warm champagne to deep bronze" becomes "cool silver to slate." **Composition variation:** Change the angle or perspective. "Flat lay, top-down" becomes "45° angled view, sitting on marble surface." **Style shift within a family:** "Matte finish" becomes "glossy finish" while keeping everything else identical. **Context change:** "On white background" becomes "on luxury tissue paper" or "in hand, holding against face." This approach saves enormous time because you're not re-explaining the product, customer, or core aesthetic—you're just testing variations on a proven brief. It's also how you build asset libraries fast.

The Troubleshooting Checklist

If an output misses the brief, run through this before re-prompting: 1. Did you name the product specifically (not just the category)? 2. Does the color palette have a hex range or natural reference, not just names? 3. Did you specify lighting direction, not just intensity? 4. Are there negative constraints ("no [X]") to prevent common mistakes? 5. Did you state the output format (3D render vs. illustration vs. mockup)? 6. Is there a style reference that isn't a generic adjective? 7. Did you describe the target customer's taste, not your taste? 8. Is the composition constraint clear ("flat lay," "lifestyle shot," "on model")? 9. Did you specify material finishes (matte, gloss, texture)? 10. Is the aspect ratio or dimension in the prompt? 11. Are there competing visual styles you need to rank-order? 12. Does the prompt avoid buzzwords that mean different things to different people? Most failed outputs fail because 3–4 of these are missing. Fill them in and regenerate.

FAQ

Do I need to use the exact same prompt formula every time?

The order matters more than the wording. Product → Customer → Reference Style → Colors → Materials → Composition → Lighting → Format. Once you have a system, you're faster and more consistent. Some designers customize the order slightly, but the five core elements always need to be in there somewhere.

How long does it take to write a good prompt using this method?

3–5 minutes if the client brief is clear. If the brief is vague, the decoding step (extracting the five elements from messy client notes) takes longer, but it's faster than guessing and revising six times. The time investment upfront prevents hours of revision loops later.

Can I use this for AI tools other than Midjourney and DALL-E?

Yes. The Master Formula structure works with Stable Diffusion, Adobe Firefly, and most other image generation tools. Tool-specific syntax varies (aspect ratio flags, model names, etc.), but the core breakdown of what to include stays the same.

What if the client's brief is genuinely one sentence?

That's when the decoding step is most valuable. Ask clarifying questions or lean on competitor references. "You like Apple's minimalism—should this packaging feel the same way, or should it stand out more because you're a smaller brand?" Good prompts come from good briefs. If the brief is thin, do the work to thicken it before writing the prompt.

Should I show the client my prompts?

You don't have to, but some designers do. It shows the client the level of detail that went into the request and why the output looks the way it does. It also gives the client language to describe what they like or don't like, which feeds back into the next iteration.

Related

AI Design Prompt Formulas →Prompt formulas for brand identity, client briefs, design revisions, and AI-assisted creative workflows. How to Keep Brand Colors Consistent When Using AI Design Tools →Stop AI from changing your approved brand colors mid-project. Learn the exact prompt structure that keeps palettes locked across all generations. How to Prevent Your Approved Logo from Morphing During Brand Guideline Updates →Keep your approved logo stable while iterating other brand elements. The revision-control prompt structure that stops unwanted AI drift in logo design. Why You're Stuck in 20+ Logo Iterations (and How to Stop) →Stop burning 20 generations per logo. Learn the revision-control prompt architecture that keeps approved elements locked while you iterate details. Brand Identity Prompt Formulas →Turn any 3–4 sentence client brief into an on-brand logo concept, color palette, and brand guideline visuals in 2–4 AI generations — not 15+....