AI Image Generation in Client Campaigns: How We Use It

The tools overlap, but each has a recognisable profile at the time of writing.

• Midjourney: strong aesthetic quality and style control; good for mood boards and atmospheric scenes.
• Adobe Firefly: trained on licensed and public domain content, built into Photoshop as Generative Fill and Expand; useful when commercial safety matters.
• OpenAI image generation in ChatGPT: good at following detailed instructions and editing an image through conversation.
• Google Imagen in Gemini and Vertex AI: strong photorealism and available through enterprise cloud contracts.
• Open models such as Stable Diffusion and Flux: can be run privately and fine-tuned on a brand’s own images.

Most campaign bottlenecks are not about a single great image. They are about producing enough good variations for every placement, market and audience segment without a new shoot each time.

Generation helps with backgrounds, seasonal versions of an existing visual, extending a photo to new aspect ratios and making regional variations. For example, a furniture retailer can show the same sofa in a Scandinavian flat and a Mediterranean terrace without shipping it anywhere.

Products get subtly altered: a seam moves, a logo gains a letter, a bottle changes shape. Generated people can look generic, and audiences have grown quick to spot that look. Copyright protection for fully generated images is limited in several countries, and some models were trained on scraped work, which raises reputational questions for brands.

Platform policies add another layer. Meta, Google and TikTok all have rules on labelling realistic AI-made or altered media, especially in political and social issue ads. Read the current policy for each platform before launch.

• Write a never-show list next to the brand guide and include it in every prompt brief.
• Use style references or a fine-tuned model rather than describing the style from scratch each time.
• Fix the palette in post-production with the brand’s exact colour values.
• Limit the number of different tools per campaign to one or two, so the look stays coherent.
• Have the same designer approve all generated assets for a client.

For variations and backgrounds, usually yes. For hero images, product detail and people the brand wants to represent authentically, a shoot often remains the better investment. Many agencies now plan one smaller shoot for core product and people images, then extend that library with generation for months.

The legal picture differs by tool and by contract. Before a generated image goes into a paid campaign, the team should be able to answer a short set of questions in writing. These are not theoretical; brand legal teams ask them, and so do some media owners.

• Does the plan we use grant commercial rights to the output, and are there revenue limits?
• Does the vendor offer indemnity against copyright claims, and under which conditions?
• Was the model trained on licensed data, scraped data or both, according to the vendor?
• Are our uploaded reference images used to train future models, and can we opt out?
• Does the output carry Content Credentials or other provenance metadata?

Adobe attaches Content Credentials to Firefly output, which records that AI was used. Some brands welcome this transparency; others need to decide how it interacts with their disclosure policy. Either way, it is better to choose deliberately than to discover the metadata after publication.

Keep the answers in the project file. When a client changes legal counsel or a campaign is audited a year later, the reasoning behind tool choices is already documented, and nobody has to reconstruct it from memory.

The answers also guide tool choice per job. A mood board that never leaves the meeting room can use whatever produces the best ideas. A national out-of-home campaign may need the tool with the clearest commercial terms, even if its output needs more retouching.

Who should run the tools?

Generation works best in the hands of people who already understand composition, light and brand rules. A designer using these tools produces usable work far faster than a non-designer with the same prompts, because they know what to fix and when to stop generating and start editing.

Train designers first, then give account managers access for quick concept sketches in client meetings, with the clear understanding that those sketches are ideas, not deliverables.