How to use a mood board with AI without copying it
Read the board for its feeling: the light, the mood, the energy, the palette and the pace. Write those down as words and send the model the words plus your own product photos. Keep logos, other brands’ products, real faces, text and named places out. Then decide, per campaign, whether the board or your brand’s own look leads.
Why the model copies the board
An image model treats every picture you send as material to use. Google’s documentation for its current Nano Banana models describes the reference slots as objects to include in the final image and characters to keep consistent. Of its three current models, only Nano Banana 2 lists a separate slot for style references, and that slot takes up to three images. Send a board photo of a surfer on a coast road and the model has good reason to put the coast road in your shot, along with the other brand’s sneakers on the surfer’s feet.
OpenAI’s prompting guide for GPT Image 2.5 gives two fixes. Give each reference a numbered role. When you want only a style, name what the reference is for (its palette, its texture or its medium) and describe the new subject separately. Adobe Firefly splits the job in its interface: a style reference guides the look and feel, and a separate composition reference guides how things are arranged in the frame.
What to take from a board, and what to leave
| Element | Take it? | How to write it |
|---|---|---|
| Light quality and direction | Yes | "Low sun from behind, long shadows toward camera, warm rim on the edges" |
| Time of day and weather | Yes | "Just after sunrise, clear sky, light haze" |
| Palette | Yes, mapped to your colors | "Bleached neutrals with one warm accent: our tea amber #C8893A" |
| Energy and pace | Yes | "Caught mid-movement, slight motion blur on hands and hair" |
| Texture and finish | Yes | "Coarse grain, faded blacks" |
| Camera angle and distance | Yes | "Low camera, subject in the lower half, lots of sky" |
| Setting | As a kind of place only | "A quiet coastal road", never the named beach |
| Logos, labels, signs | No | List them under "keep out" |
| Other brands’ products | No | List them under "keep out" |
| Real people’s faces | No | Cast your own model |
| Text in the images | No | Leave it out |
| Named places, landmarks, artworks | No | Describe the kind of place instead |
Write the board down as words
Before any render, the art director reads the board and writes a board read: about ten lines that turn the feeling into instructions a model can follow. The read goes into the prompts, and the board images stay with the team. For words that describe light, see lighting words that work in AI prompts.
BOARD READ Campaign: ________ Board: __ images Date: ________ THE ONE IDEA (one sentence: what every image shares): ________ LIGHT: ________ TIME AND WEATHER: ________ PALETTE, MAPPED TO OUR COLORS (with hex): ________ ENERGY (still / slow / mid-movement / chaotic): ________ TEXTURE AND FINISH: ________ CAMERA ANGLE AND DISTANCE: ________ SETTING, AS A KIND OF PLACE: ________ KEEP OUT: logos, other brands’ products, real faces, text, named places, ________ WHO LEADS (brand look or board), AND WHY: ________
A worked example
A made-up board of nine images for a summer launch by the made-up tea brand from the visual DNA guide: surfers walking out at dawn, a blurred cyclist, sun on a white wall, a faded travel poster, salt drying on skin, a café terrace before opening, a swimmer’s wet hair, a paper cup on a dashboard, a quiet coastal road. The read:
THE ONE IDEA: early heat before the crowd arrives. LIGHT: low sun from behind the subject, long shadows toward camera, warm rim light on edges. TIME AND WEATHER: just after sunrise, clear sky, slight haze. PALETTE: bleached neutrals with one warm accent. Chalk #F2EEE6 and tea amber #C8893A. ENERGY: mid-movement, a little motion blur on hands and hair. TEXTURE AND FINISH: coarse grain, faded blacks. CAMERA ANGLE AND DISTANCE: low camera, subject in the lower half, lots of sky. SETTING: a quiet coastal road and a café terrace, as kinds of place. KEEP OUT: the surfboard brand in image 1, the poster’s lettering, the cyclist’s face, the logo on the paper cup. WHO LEADS: the board, inside the brand palette and never-do list, because this is a launch story.
Nothing in the read names a product, a brand or a person from the board. Every line describes light, time, color, movement or framing, which the model can apply to your can without copying anyone’s picture.
How many images should a board have?
Enough to show one idea in several forms, and few enough that the idea stays clear. If you need more than a dozen images to explain it, the board probably holds two ideas, so split it into two reads. When you do send images, the caps are small: Nano Banana 2 accepts up to three style references in one request. A written read has no cap.
When should the board lead, and when should the brand?
| Job | Who leads | Why |
|---|---|---|
| Product page and catalog photos | Brand | Shoppers compare these side by side, so they must match each other and last season’s photos |
| Always-on social posts and ads | Brand, with the board as seasoning | People recognize the same look across months of posts |
| Campaign launch or seasonal story | Board, inside the brand palette and never-do list | The point is a new feeling, and the palette keeps it yours |
| Collaboration or limited drop | Board | The partner’s world is part of the product |
| Pitching routes to a client | Board | Each route exists to show a different feeling |
Whoever leads, the product never changes. Its shape, label and color come from your product photos, and the board has no say over them. Keep the never-do list in force either way.
Questions people also ask
- Can I put competitors’ ads on the board?
- Keep them for the team’s eyes only. Do not send them to the model or name the competitor in a prompt, because the model can reproduce their products, layouts or logos. Write down what works in them as light, pace or palette.
- Should we make the mood board with AI?
- You can. It sidesteps questions about other people’s photos, and it lets you test a feeling on your own product before the campaign. Treat an AI-made board like any other: write the read, then decide how much of it the campaign keeps.
- What if the model still copies the board?
- Take the board images out of the request and send only the written read plus your product photos. If one image must go in as a style reference, crop it to the part that carries the look, such as a patch of light on a wall or a texture, so there is no object in it to copy.
Where Overs fits
Overs takes a mood board of up to twelve images plus a note. It reads the light, mood and energy of the board and applies that feeling to your product in your brand’s style instead of copying what is in the pictures. A dial sets whether the brand’s own look or the board leads.
Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.
Sources
- Google AI for Developers: Nano Banana image generation (reference image types and limits)
- OpenAI: Image prompting, GPT Image 2.5 prompting guide (assign roles to references, transfer a visual style)
- Adobe Help Center: Style reference images in Adobe Firefly
- Adobe Help Center: Match image composition to reference image in Firefly