How do we keep AI-generated campaign images on brand?

Keep AI-generated campaign images on brand through human creative direction, image references and repeated testing against the intended style. Depending on the tool, that can include adjusting prompts and settings or training it to mimic a specific style. David Alexander's account describes an iterative process, not guaranteed brand consistency from a single prompt.

How quickly can we create AI-generated campaign images?

An initial AI-generated campaign image can be produced in minutes, but that is not a reliable estimate for a finished, approved asset. For UST's AI services microsite, The Frameworks used Midjourney to generate healthcare and manufacturing scenarios quickly. The accompanying process account identifies prompting, testing and refinement as additional work, so production planning should distinguish initial generation from campaign readiness.

How much refinement do AI-generated campaign images need?

AI-generated campaign images can require substantial prompting and refinement, but there is no established standard number of review rounds in the described workflow. Refinement may involve settings, style training, image references and repeated testing, rather than prompt changes alone. David Alexander describes reaching the intended image as a creative skill that requires thought and practice.

When should we use AI-generated images for our campaigns?

Choose the production route that fits the image requirement

AI generation is worth considering when the required scene is difficult to source and does not depend on photographing actual people or places. For UST's AI services microsite, The Frameworks used Midjourney to create realistic healthcare and manufacturing scenarios that were difficult to demonstrate through existing imagery. That is a practical example of AI supporting a specific creative requirement, not proof that AI is always preferable to photography.

Stock remains useful when a suitable asset exists, although finding it can require substantial searching. Adapting a licensed image offers another route when the starting asset is close to the intended result. Original photography remains relevant when the campaign needs real subjects: photographing employees, for example, produces imagery specific to the organisation.

Compare these routes against the same visual requirement. A hard-to-find imagined scenario and an authentic portrait present different production problems, so they should not automatically receive the same solution.

Sources: Powered by imagination: how I’m making stock imagery creative with AI, How to deliver cracking video content during lockdown, The inside story: why your people shouldn't be just passengers on your brand journey

Brand consistency requires creative direction and refinement

AI-generated campaign imagery needs a defined visual target and human judgement throughout production. Prompts are only part of that work. Depending on the tool, refinement can involve changing settings, supplying image references, training for a particular style, and testing the resulting images against the intended look.

David Alexander, Creative Director and Head of Studio at The Frameworks, describes this as "co-piloting": creating an image in a brand's style still requires creativity, practice and substantial prompting. The account does not establish a standard number of review rounds or guarantee consistency across a campaign.

Independent research supports the narrower value of iteration. A study published in the 2025 IJCAI proceedings found that human prompt adjustments improved alignment with target images in both human assessments and quantitative measures. That finding supports a directed refinement process, but it does not establish that a tool will reliably follow every brand photography guideline.

Sources: Powered by imagination: how I’m making stock imagery creative with AI

Judge the finished campaign asset, not generation speed alone

A fast image-generation stage does not establish the total cost or commercial suitability of a campaign asset. Compare production routes using the full workload: searching or prompting, selection, retouching, revisions, legal review and stakeholder approvals. The Frameworks' UST account describes rapid generation, while David Alexander's process account makes clear that reaching the intended style still requires testing and refinement. Neither provides a complete campaign cost benchmark.

Rights also need a separate decision. The US Copyright Office concluded that human-determined expressive elements, including creative arrangements or modifications, can receive copyright protection, but prompts alone do not establish that protection. Copyright protection is not the same question as permission to use an image in advertising.

Finally, visual quality should not be treated as proof of audience performance. The 2025 working paper AI in Disguise found a click-through advantage only when generated images did not appear AI-made. Research in the Journal of Advertising Research found more negative responses to disclosed AI imagery in its first luxury-advertising study. Those findings support testing the proposed treatment rather than assuming a universal advantage.