What to Type for AI

Why You're Stuck in 20+ Logo Iterations (and How to Stop)

You approve a logo direction. Then you ask the AI to refine the color. Suddenly the icon shifts. You lock it down again. Now the letterform looks wrong. Three hours later, you're back where you started.

This isn't a tool problem. It's a prompt architecture problem. When you don't structure which parts of your brief are locked versus flexible, the AI regenerates everything on every request—including the parts you already paid for.

There's a better way: a three-layer prompt structure (LOCKED/VARIABLE/CONSTRAINTS) that holds approved elements rigid while you iterate the parts that actually need work. Most freelancers who implement this see average iteration counts drop from 14–20 generations down to 2–3.

The Problem: Generic Prompts Regenerate Everything

A typical logo brief looks like this: "Modern tech logo, geometric, minimal, professional blue." When you send that to Midjourney or DALL-E, the AI treats every attribute equally. You get a result. You ask for "more refined geometry." The AI rebuilds the whole thing—new icon, new color, new proportions. Then you ask for "keep the icon, change the blue." But the AI doesn't remember what "the icon" was supposed to be. It regenerates again. After 15–20 cycles, you've burned client patience and your own time, and you still might not have nailed it. The root cause: your prompt has no hierarchy. The AI doesn't know which decisions are final and which are still in play.

The Solution: Lock, Vary, Constrain

The LOCKED/VARIABLE/CONSTRAINTS structure fixes this by telling the AI exactly which parts are fixed and which are negotiable. **LOCKED:** Describe the elements already approved by the client. "Icon: geometric circle with negative space cut-out (as generated in round 3, exact form)". **VARIABLE:** Name what you're actually iterating. "Color palette: explore blues between navy and teal, maintain 60/40 contrast ratio." **CONSTRAINTS:** Set universal rules that apply to everything. "All assets maintain 2px minimum stroke width. No gradients. Helvetica-weight typography only." When the AI sees this three-layer structure, it stops regenerating the locked parts and focuses compute on the variable space. You get consistency. You reduce wasted iterations to the parts that actually need refinement.

Real Example: 4 Generations Instead of 15

Client brief: fintech logo with mark + wordmark, professional but approachable. **Generation 1** (initial direction): You send the full brief with tone codes (B2B fintech: "trust signals, modern without cold, regulated industry sensibility"). Get 4 logo directions. **Generation 2** (client picks direction 2): You lock the icon form. Vary the wordmark treatment and explore 3 sans-serif weights. Client approves wordmark. **Generation 3** (client requests color): You lock both icon and wordmark. Vary only the color—specify the approved icon using the LOCKED layer so the AI doesn't rebuild it. Client approves teal. **Generation 4** (final refinement): Lock icon, wordmark, and color. Vary only the size ratio and padding rules for the guideline cover version. Done. Total: 4 focused generations. Without the structure, you'd iterate the icon 8 times, the wordmark 4 times, and the color 3 times before nailing it.

How to Extract What to Lock From Your Brief

The question isn't "what should I lock?"—it's "what did the client already approve?" Use this checklist: • **Icon/mark**: If the client said "yes" to a direction, that goes in LOCKED. • **Color direction**: If they approved "navy and teal" but not the exact hex values, that's VARIABLE (let the AI explore within the range) with a CONSTRAINT (maintain the blue-to-teal ratio). • **Typography**: If they approved a serif or sans-serif choice, lock it. If they're still deciding weight, vary it. • **Tone/feeling**: Always goes in LOCKED via the tone code bank—this prevents the AI from drifting toward generic wellness-speak or startup-cliché when you're iterating details. The prompt formulas included in this product walk you through this extraction for logos, palettes, typography specimens, patterns, and full guideline sets. Each one includes a worked example so you see exactly which variables came from the brief.

Why Iteration Counts Drop So Fast

Every wasted generation is the AI doing unnecessary work. It's rebuilding approved parts because the prompt didn't tell it not to. With LOCKED/VARIABLE/CONSTRAINTS, you're not reducing the AI's thinking—you're redirecting it. Instead of regenerating the icon 8 times while you hunt for the right blue, the AI explores blue space while keeping the icon stable. Instead of watching the wordmark shift every time you adjust spacing, you lock it and iterate spacing alone. The formulas also include cliché exclusion rules for your industry (fintech, SaaS, wellness, food, luxury). When you lock in a tone code + exclusion set, you don't get a second generic attempt—you get a genuinely differentiated iteration on the first try.

Syntax You Can Copy and Paste Today

Both Midjourney and DALL-E 3 respond to clear structural language. The product includes copy-pasteable prompt templates for each formula and tool—no translation needed. You'll see things like: ``` LOCKED: Icon geometry [specific description from approved round] VARIABLE: Color exploration within [range] CONSTRAINTS: Stroke width minimum 2px, no anti-aliasing softness, typography weight [locked value] ``` Just fill in the brackets from your brief, and the AI treats it as instruction, not suggestion. This works for quick 4-sentence briefs and complex brand guidelines alike.

FAQ

Does this work for both Midjourney and DALL-E 3?

Yes. The LOCKED/VARIABLE/CONSTRAINTS structure translates to both. The product includes syntax examples for each tool so you're not guessing which phrasing works where.

What if the client keeps changing their mind about the locked parts?

Then they move from LOCKED to VARIABLE. You'll iterate that section again, but now you know exactly how many generations it'll take because you're only varying one layer instead of rebuilding the whole brief.

Do I need to use all 10 formulas, or just the ones relevant to my project?

Use what you need. A quick logo-only project might use Formula 1 + tone codes. A full brand guidelines build would use Formulas 4, 5, 6 in sequence. The worked examples show how they stack.

How do I know if my prompt is too rigid and blocking good ideas?

If you're stuck after 3 iterations on the VARIABLE layer, move something from LOCKED to VARIABLE and regenerate. The goal is fast client approval, not perfect first-generation output. The structure just prevents wasted regeneration of approved work.

Can I use this for brand updates or refreshes, or just new identity builds?

Both. For refreshes, your LOCKED layer is the existing brand elements you're keeping. Your VARIABLE layer is what you're updating. Same structure, same iteration efficiency.

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. How to Write AI Image Prompts From Vague Client Feedback →Convert 'it feels off' into precise prompt parameters. The 3-step framework graphic designers use to nail revisions in one round instead of seven. 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+....