How to make on-brand ad creative with AI

Why ad sets drift

A regenerated variant can change more than its background: product details, lettering and logos can move too. Variants built from earlier variants compound those changes, while resizing through generation creates another composition to check.

Ad platforms can add their own backgrounds, copy or videos. Include those outputs in approval.

Decide what stays fixed

Put exact elements into a layout your agent exports; generate the scene around them. Video creative follows the same split.

Write the ad brief

Specify audience/angle, approved offer and dates, placements/safe zones, exact image and platform-field copy, claim sources, references and the one thing each round varies.

Rainbrand maintains the shared voice, palette, fonts, assets and claim guidance. A Brief selects what the campaign needs. Describe its task-specific copy/placements and keep current product references in the Brand Library.

Resolve missing prices or unsupported claims before generating.

Get the sizes right

Use current platform specs. In the October 2026 Meta and Google sources:

• Safe zones: Meta’s Stories and Reels guidance leaves roughly 14% top, 35% bottom and 6% sides free of key elements.
• Model dimensions: GPT Image 2.5 requires multiples of 16, so generate at a supported larger size with the same ratio, then crop/scale to the final size.
• Separate assets: Responsive display assembles images, text and logos. Performance Max recommends a clean image per ratio and content within the centre 80%.
• White backgrounds: avoid a white transparent logo that disappears when the platform places it on white.

Performance Max headlines are limited to 30 characters. Decide which words belong in platform fields and which in the picture.

Keep the product real

FTC guidance assesses pictures as well as words. Invented features or altered packaging can misrepresent the product.

For Eager, a juice brand, Claude used Rainbrand to make a blackcurrant-flavour campaign concept. A carton’s lettering used the wrong type/weight; Review flagged it and Claude reset the type:

The next typography check passed. This is one image, not a certified resized set; inspect the corrected master’s lettering again in each placement. The full example keeps the findings.

Set the words last

Place the original logo and exact copy in the layout, using the approved fonts and clear space. This keeps offers editable and avoids asking a model to redraw the mark.

If text belongs inside the generated scene, follow OpenAI’s exact-wording guidance and Google’s advice to settle copy first. Read every letter at the final size. Voice-guide instructions cover headline rules.

Vary one thing at a time

1. Use one completed campaign Brief with each supported AI Gateway request; Rainbrand adds guidance/assets. With other tools, supply that context yourself.
2. Branch from approved references and the master, not the last generated variant.
3. Crop/re-layout when sufficient; generation for a new ratio needs another check.
4. Retain prompts, references, model and settings for each approved variant.

The campaign image-set guide shows separate 1:1 and 16:9 generations from a 4:5 master.

Platform settings matter:

• Meta enables generation/expansion per ad and requires previewing generated creative.
• Google may enhance assets; Final URL expansion can generate text, and missing video orientations can trigger generated videos.
• Check inferred Performance Max brand guidelines against the actual brand.

Request a Rainbrand Brief for [your brand] for the attached ad campaign. Use its approved offer, exact copy, audience, claims and placement requirements. Flag missing or conflicting facts before generating. Make background options without text, logos or invented product details and show them to me. Once I choose one, adapt it to the required sizes and safe areas. Place the approved product image unchanged, set the exact copy in the supplied fonts and use the original logo file. Export each final placement, keeping clean images where the ad platform needs them. Request a Review of each finished image with its expected copy and show me any uncertainty. Wait for approval before publication.

Check every variant before launch

Use Rainbrand Review on each final image with expected copy and relevant exceptions, such as intentionally different packaging colour. Revise confirmed failures and keep uncertainty for the decision-maker.

Save the Brief with variants. If the brand changes, decide which ads need updating and retrieve a fresh Brief before revision.

Google’s image guidance describes AI-label settings and market rules; the client-approval guide covers disclosure. Start with one existing campaign using the image-set workflow.