TL;DR
Don’t ask AI for “a brand look” — ask it for 3 genuinely different directions so the choice is real. The prompt below returns, for each direction, a palette (5 hex codes with roles), a real type pairing, a photography/illustration style, a logo vibe, a ready-to-paste image prompt, and the one place each direction would fail. Run the image prompts in Midjourney V8.1 or ChatGPT’s image tool, pin all three at 3 metres, then pick. Treat AI as the explorer, not the decider.
The task
You are starting a brand from zero — landing page, social profile, packaging, deck — and you need a visual direction consistent enough that strangers recognise it. The mistake at this stage is committing to “the brand look” before you have explored. A single direction sounds decisive, but you are choosing between option A and nothing. The job is to generate three genuinely different directions so the choice is real.
This is an exploration prompt, not a logo generator. It is meant to be run early, before you book a designer, so you walk into that conversation with a chosen direction instead of a blank page.
When AI helps — and when it does not
AI is good at synthesising mood, generating coherent palettes, suggesting type pairings, and writing image prompts that produce on-brand visuals across tools. It is poor at judging whether a direction fits — that needs taste, customer reaction, and competitive context. Use AI to explore wide; bring a human in to choose.
One concrete limit worth knowing: AI hallucinates font names. Always verify every typeface on Google Fonts (about 1,900 families as of June 2026, all under the SIL Open Font License — free for commercial use, including logos) or in Adobe Fonts before you build with it.
What to feed the AI
- Brand name and one-line meaning (“Folio = the page where craft becomes work”)
- Audience, and the brands they already trust
- Brand values in five words (“calm, precise, modern, opinionated, kind”)
- Mood references — adjectives, places, eras, art movements
- No-go visuals — competitor looks, dated aesthetics, anything off-brand
- Output formats you will need (web, app icon, print, packaging)
The richer the input, the less the model defaults to generic startup-blue. Vague in, vague out.
Copy-ready prompt
Propose 3 distinct brand visual directions.
Brand: [name]
Meaning in one line: [line]
Audience and trusted brand references: [list]
5-word values: [list]
Mood references: [adjectives, places, eras]
No-go visuals: [list]
Output formats needed: [web / app icon / print / packaging]
For each of 3 directions, return:
1. Direction name (2-3 words) + 1-sentence description
2. Colour palette: 5 hex codes (primary, secondary, accent,
neutral light, neutral dark) with role notes + a dark-mode
equivalent for each hex
3. Type pairing: heading typeface, body typeface, fallback stack
(use only real Google Fonts or Adobe Fonts — no invented names)
4. Photography / illustration style — 2 sentences
5. Logo lock-up vibe in words
6. A representative image prompt I can paste into Midjourney or
ChatGPT's image tool to anchor this direction
7. Where this direction would fail (audience, format, or
competitor adjacency)
The three directions must be visually different. If two are too
close, kill one and create a contrast. Verify every typeface
exists before listing it.Variant for fast-moving categories: add the line “One direction should feel deliberately ‘wrong for the category’, to test whether contrast is more memorable than fit.”
Which model to run this in (June 2026)
The prompt is plain text, so any frontier chat model handles it. The differences show up in the image step.
| Tool | Best for | Notes (as of June 2026) |
|---|---|---|
| ChatGPT (GPT-5.5) | Drafting the 3 directions + generating images in one thread | Image generation is native (GPT Image; DALL-E 3 was retired May 12, 2026). Plus is $20/mo, Free is $0 with tight limits. |
| Claude (Sonnet 4.6 / Opus 4.7) | The strongest written direction reasoning | Text only — no image generation. Pair with a separate image tool. Pro $20/mo. |
| Gemini 3.1 Pro | Direction text + image generation | Google AI Pro $19.99/mo; 1M-token context. |
| Midjourney V8.1 | Highest-fidelity anchor images | Released Apr 30 2026; native 2K, --cref for character/style consistency. Plans $10 / $30 / $60 / $120 per month — no free tier. Put literal text in "quotes" so it renders. |
Practical split most teams use: draft the three directions in ChatGPT or Claude, then run the image prompts in Midjourney for the highest-quality anchor frames.
Recommended output structure
A side-by-side comparison: 3 columns by 7 rows. Print it, or paste it into Figma and pin three boards on a wall. The test is physical — you should be able to walk past at three metres and tell which is which. If you can’t, two directions are too close and one should be replaced.
How to check the output is usable
- The three palettes do not share a primary hue family (warm vs cool vs neutral)
- Type pairings are real, available pairings on Google Fonts or Adobe Fonts — not invented font names
- Each direction has a stated failure mode. If AI says every direction is great, push back and demand a weakness
- The image prompts produce consistent images when run three times each (in Midjourney, lock this with
--crefor a style reference) - A customer in your target audience would describe each direction with a different one-word reaction
Common mistakes
- Single direction = no comparison = no choice
- Asking for “the logo” instead of “directions”. See the logo direction prompt
- Letting AI invent font names that do not exist. Verify every typeface on Google Fonts
- Picking the direction that “feels safest”. Safe usually means it looks like the category leader
- Skipping the image prompts. Without a representative image, a “direction” is just a paragraph
- Forgetting dark mode. Ask for a dark-mode equivalent of every hex from the start, not as a retrofit
FAQ
- Can AI generate the final logo? No. AI is great for exploration, but a committed logo still benefits from a designer who handles vector cleanup, accessibility contrast, and how the mark behaves at favicon size versus billboard size.
- How many colours is too many? Three directions give you roughly 15 candidate colours. Pick 5 to 7 for the final palette: one primary, one or two secondaries, one accent, plus light and dark neutrals.
- Midjourney or ChatGPT for the images? ChatGPT (GPT Image) is faster, free-to-cheap, and keeps text and images in one thread. Midjourney V8.1 gives higher fidelity and tighter style consistency via
--cref, but starts at $10/mo with no free tier. Draft in ChatGPT, finalise anchors in Midjourney. - Are Google Fonts safe to use commercially? Yes. Every family is open-source (usually SIL OFL), free to use in websites, apps, print, and logos. Adobe Fonts are licensed through a Creative Cloud subscription instead.
- What about dark mode? Build it in from the start. The prompt above already asks AI to specify a dark-mode equivalent for every hex, so you don’t rebuild the palette later.
Related
- Brand voice definition prompts — the verbal counterpart to visual direction
- Brand positioning statement prompts — positioning that the visuals will express
- AI brand visual direction — full workflow including image generation
- Logo direction prompt — narrow down from style to mark
- Moodboard prompt — build a reference moodboard first
- Visual style consistency prompt — keep the chosen direction consistent across assets
Tags: #AI writing #Workflow