TL;DR
Feed an AI image model a tight concept brief (mood words, hex palette, one style anchor, a variety axis) and ask for 9 to 16 prompts that share one look. Generate them in a single session, lay them out 3x3, and keep the “style spine” line to reuse on every future image. For consistency across the set, use Midjourney’s Moodboard or Style Reference, Gemini’s Nano Banana Pro for face/portrait consistency, or GPT-Image-2 when the board carries on-image text. The copy-ready prompt below works with any of them.
The task
You are pitching a creative direction — a brand refresh, a campaign concept, a product-launch identity, an interior proposal — and you need a moodboard that communicates the feeling in seconds. You want 9 to 16 images that read as one project, generated in a single sitting so cost and time stay low.
The hard part is not generating pretty pictures. It is generating nine different pictures that still look like siblings: same light, same palette, same grain, different subjects.
Which model to use (June 2026)
The prompt template below is model-agnostic, but the consistency mechanism differs. Pick by what your board needs most.
| Model | Best for moodboards when | Consistency tool | Cost (June 2026) |
|---|---|---|---|
| Midjourney V7 / V8.1 | Aesthetic-led brand/editorial boards | Moodboards (--p mID) + Style Reference (--sref) | $10 Basic / $30 Standard / $60 Pro / $120 Mega per month |
| Gemini Nano Banana Pro | Boards with recurring people, faces, products | Multi-reference image input | Google AI Pro $19.99/mo; API metered |
| GPT-Image-2 (ChatGPT) | Boards that need legible on-image text or layout | Reference image + chat memory | Bundled with ChatGPT Plus $20/mo |
| Flux 2 | Photoreal boards on a tight per-image budget | Style/reference conditioning | ~$0.01–$0.10 per image on hosted APIs |
Notes: GPT-Image-2 shipped April 21, 2026 and topped the LMArena Image leaderboard at launch (Elo ~1512), with the strongest text rendering and layout. Nano Banana 2 / Pro (Google) remains the leader for face consistency and multi-person compositions. Midjourney V7 added a dedicated Moodboards personalization feature: build a profile from reference images, then append its code (--p mID) to every prompt so the whole set inherits one aesthetic. On Midjourney, run early exploration in Draft Mode (roughly 10x faster and about half the GPU cost) before promoting winners to full quality.
When AI is the right tool
- Early-stage concepting where speed matters more than final-pixel polish.
- Internal alignment between strategy, design, and stakeholders before commissioning a photographer.
- Iterating across multiple visual directions for a client to react to.
- Solo founders and small teams without a design budget.
When not to rely on AI alone
- Final deliverables. A moodboard is a starting point, not the campaign.
- Anything involving real people, recognizable public figures, or specific brand logos.
- Technical product renders that must match exact CAD specs.
- Categories where image-model bias is documented (skin tones, body types, professional roles) — review the set before it leaves the room.
What to feed the AI
- Theme or concept in one sentence.
- 3 to 5 mood words (e.g. “quiet, optimistic, sunlit, lived-in”). More than five and the model averages them into nothing distinctive.
- Color palette as hex codes or named references (“burnt sienna, oat, charcoal”).
- One or two style anchors — real reference photographers, films, or art movements.
- The variety axis — what should differ across the nine (subject, time of day, scale).
- Aspect ratio and final use (slide deck, website hero, print). On Midjourney, set
--arbefore writing the prompt; it changes how V7 composes, not just how it crops.
Copy-ready prompt
You are a creative director helping me build a 9-image moodboard.
Concept: [concept_one_liner]
Mood words: [mood_words]
Palette: [palette]
Style anchors: [reference_photographers_or_films]
Variety axis: [subject_or_setting_variations]
Aspect ratio: [ratio]
Output format: 9 image prompts for [image_model].
Rules for each prompt:
- 1-2 sentences, present tense.
- Always include lighting (e.g. "soft late-afternoon", "overcast diffused").
- Always include a camera or lens cue ("35mm film", "medium format portrait").
- Vary subject and composition across the 9, but keep palette and lighting consistent.
- No people with identifiable faces unless I confirm in advance.
- End each prompt with the palette anchor line.
Output as a numbered list. After the list, summarize the shared style spine in 2 sentences.Replace each [placeholder] with your own values. The final line — “summarize the shared style spine” — is the part that earns its keep: that two-sentence spine is what you reuse everywhere afterward.
Recommended output structure
A numbered list of 9 image prompts plus a short “style spine” paragraph. Paste the spine into any future image prompt for this project to hold the look. On Midjourney, the durable version of the spine is your Moodboard code (--p mID) or a --sref code; on Gemini or ChatGPT, it is a reference image plus the spine text in the chat.
How to check the output
- Generate all 9, lay them out 3x3, and squint. If one jumps out as off-style, regenerate it with the style spine reinforced.
- Verify palette adherence — drop the grid into a color sampler and compare against your hex list.
- Check subject variety. Nine portraits of the same face is not a moodboard.
- Show it to someone outside the project and ask them to describe the mood in three words. If they cannot, the board is not communicating.
Common mistakes
- Style drift — every prompt invents its own lighting. Fix: name the exact light once and repeat it verbatim.
- Repetition — the same subject nine times. Fix: write an explicit variety axis.
- Mood-word soup — six adjectives that cancel each other out. Cap it at five.
- Thumbnail blindness — generating at low resolution and missing detail problems until print. Upscale before you judge.
FAQ
- How many images make a moodboard? Nine (3x3) is the standard grid. Use 12 to 16 for richer pitches or print spreads.
- Which model is most consistent across a set? Midjourney Moodboards/Style Reference for aesthetic cohesion; Nano Banana Pro when the same person or product must recur; GPT-Image-2 when on-image text must stay legible.
- Should I put text on the images? No. Text belongs on the deck around the moodboard, not baked into the visuals.
- Can I mix AI and stock images? Yes. A moodboard is about feeling, not authorship.
- What if the client picks one image to actually use? Replace it with a real photo or a licensed shot. Moodboards are direction, not deliverables.
- How do I keep cost down? On Midjourney, explore in Draft Mode (about half the GPU cost) and only upscale the keepers; on Flux 2, hosted generation runs roughly $0.01–$0.10 per image.
Related
Pair this with brand voice definition prompts, brand positioning statement prompts, and brand style prompts. For background plates, see app background prompts. Midjourney’s own Moodboards documentation covers the --p profile workflow in detail.
Tags: #AI writing #Workflow