How to apply your brand guidelines to AI-generated images

Start with the image's job

“Make an image in our brand style” leaves almost every useful decision open. The model still has to choose a subject, camera angle, lighting, layout and relationship between the product and the message. A list of colours cannot make those choices for it.

Write down what the image needs to accomplish before turning the brand guide into a prompt. A website hero explaining a product needs different information from a social post announcing an event. State the audience, message, destination, dimensions, required copy and things the picture must not imply.

For Acme, the task might be to introduce Atlas Pro's shared canvas to design leads in a landscape website hero. The product interface should be recognisable, with room for the headline on the left. The image must show the actual product rather than suggest a feature that has not shipped. That gives you a basis for selecting guidance and files.

Say which instructions apply to this image and which apply to all brand work. “Leave the left third open” is a layout choice for this image. “Never stretch the logo” is a standing rule.

Turn the guide into decisions

Read the parts of the guide that change what this picture should show. Record each relevant rule and the reference that demonstrates it.

Explain roles as well as values. “Use burgundy for one focal accent against a pale ground” gives direction that a palette alone does not. Distinguish an inspiration image from an asset that must be preserved. A lighting reference shows the lighting to follow; its product and typography may have nothing to do with your brand.

When the guide offers no answer, make the creative choice explicit in the brief. For example, a new camera angle can be a proposal for this image without becoming a rule for every campaign. Leave irrelevant sections out.

Choose inputs the model can use

Check the selected tool's current support for reference images, editing, masks and output dimensions. Google's image-generation guide documents image editing and combining references. OpenAI's guide covers generation and editing, while noting remaining difficulties with precise text placement, structured compositions and consistent brand elements.

Assign each reference a job. For a product-led image, supply the approved product view. For an illustration, supply an approved illustration and explain which characteristics to follow. If references differ, identify which one controls the product, which shows the lighting and which demonstrates the layout.

Use a record like this to make the inputs understandable to the next agent or designer:

Task and destination:
Dimensions and required crops:
Message and exact copy:
Brand rules relevant to this image:
Reference files and the purpose of each:
Original assets to preserve:
New creative choices for this task:
What to generate:
What to compose or typeset afterward:

The record should point to files the tool can access. A filename in a prompt is only useful if the file is also available to the tool making the image.

Preserve the parts that must match

If exact reproduction is required, place the supplied file in the final composition. This is especially useful for logos, interface screenshots and product details that must be recognisable. Set required text in the actual font when the layout needs precise typography or later edits.

William & Mary's AI brand guidance makes this explicit for its own identity: use official assets and avoid approximate logos, colours or fonts. Your own guidelines determine which elements need that treatment in your work.

For the fictional Acme hero, a practical division would be:

• Generate a background with the approved lighting, materials and room for the product.
• Place an approved Atlas Pro screenshot at its intended size and crop.
• Typeset the headline in the page or layout tool.
• Add the original Acme logo and inspect its clear space.

For Eager, a juice brand, we asked Claude to develop a campaign for a blackcurrant-flavour concept. The landing page places an archived SVG wordmark, rather than relying on a generated approximation. Its saved stylesheet loads actual font files for the page text.

The page uses the supplied mark and type directly. The Eager example shows the resulting work and its recorded checks.

Give agents current brand context

For recurring visual work, the same rules and files need to be available across campaigns and generation tools. Rainbrand maintains brand context that connected agents can retrieve for those tasks. A Brief selects the guidance and assets relevant to the picture you describe.

Request it with the audience, destination and constraints. Then pass the original task, the completed Brief's guidance and its relevant assets to your existing generation tools. If you use a supported model through Rainbrand's optional AI Gateway, supply the completed Brief with the generation request. Merely associating the request with a brand does not supply that context.

Request a Brief for Acme's Atlas Pro website hero, for design leads. We need a landscape composition introducing the shared canvas, with space for a headline on the left. Show me the relevant image direction and selected assets. Separate references to imitate from files to preserve. Propose what to generate and what to place as original assets before making the image.

The same sequence can run inside a creative product or service: your workflow requests the guidance, creates the work and handles revisions. The campaign image guide shows how to carry a selected direction into several formats.

Change one thing at a time

Check the first image against the rules before asking for more variations. Name the specific error: a wrong product detail, too much accent colour, insufficient open space or a reference used for the wrong purpose.

Give a revision instruction that also preserves what is working. “Keep the product view and lighting; remove the decorative objects and leave more space above it” gives the tool a change to make and details to retain. If the product itself is wrong, return to the product reference before polishing the background.

Once you choose a direction, use the selected image and the same guidance when adapting formats. A landscape crop may need a new composition for a portrait placement. Check the subject, copy and hierarchy again after the change.

Inspect the finished composition

The final image is the one after typesetting, logo placement and cropping. Check that version at the size and destination where it will appear.

Rainbrand Review checks work against the brand. For this workflow, give the agent the exported image or a visual capture of the finished composition, along with the expected copy and relevant constraints. Ask for the findings and available evidence pictures. The agent uses its editing tools to make corrections, then submits the revised image for another Review.

Keep the checked composition, its original files and the brief together. When you make the next format, start from those inputs and inspect the new export again.