Production Guide·7 min read·October 26, 2026

How to Brief AI-Generated Campaign Visuals

A practical brief for deciding what can be generated, what must stay product-accurate, and how to review AI-assisted campaign images before release.

Illustrative production desk with an unbranded physical product, moodboards and a matching draft campaign visual on a monitor.
Illustrative image: define what may change creatively and what must remain true to the product.

A useful brief for AI-generated campaign visuals starts by separating creative freedom from facts that cannot change. State what the image must communicate, where it will run, which elements may be invented or transformed, and which product, person or place details must remain accurate. Then define how the team will check the result. Without those boundaries, a polished image can still be unusable as an advertisement.

AI generation can be a way to explore art direction, make an imagined environment or extend an approved visual system. It is not automatically the right method for every scene. A product with precise packaging, a real spokesperson or a location that must be represented faithfully may call for photography, filming, design, compositing or a hybrid approach. Visual Matters' AI visual generation service frames this as a choice made from the brief, with human art direction and review guiding the work.

1. Decide the production method before choosing a tool

Write one sentence about the job of the image: “The audience should understand…” Then list the assets and references already available. If you have approved pack photography, a filmed product demonstration or a brand illustration system, those may be better starting points than generating the whole frame. If the setting is imaginary but the product must be exact, a hybrid image can keep the verified product asset while using generation for an environment or exploratory background.

Ask three questions at the outset. Does the final image need to depict a real item or event accurately? Would a mistake be difficult to detect at normal viewing size? Is the source material available and cleared for this use? A “yes” to the first two is a reason to increase control, not an automatic ban on AI. It may mean using a real photograph as the product layer, restricting generation to less factual elements, or choosing traditional production for the whole image.

2. Define what is fixed and what can vary

Create two columns in the brief. The fixed column might contain product shape, logo, label copy, approved claims, colour, proportions and required accessories. The flexible column might contain lighting mood, background material, camera angle, graphic treatment or season. Add a third column for unresolved questions, such as whether a proposed setting implies a product use that has not been approved.

Provide current reference assets, not screenshots passed through several rounds of messaging. Name the version of the pack or product that should appear. Decide who checks each fixed element: a product owner for details, a brand lead for identity, and an appropriate specialist for regulated claims. A creative director can judge the image's strength but should not be the sole authority on a factual product specification.

Google's guidance on generated images in Google Ads explicitly tells advertisers to review generated assets for accuracy, misleading content and policy compliance before publication. That is a useful publishing principle even when the image is made outside Google Ads. An attractive draft is a proposal, not evidence that a product, person or claim has been represented correctly.

3. Specify the intended placements and review size

List the planned uses before generation: website hero, paid social, store display, email, print or presentation. Record each required aspect ratio and any space needed for copy, product information or a call to action. A single wide image may not crop well into a vertical placement. Decide whether a second composition is required, rather than treating resizing as a purely technical export.

Review drafts at the size and in the layout in which people will see them. Enlarged on a studio monitor, a subtle label error may be obvious; on a phone, a distorted product shape may be harder to notice but still misrepresent the item. Conversely, an intricate generated background can look compelling full-screen and turn into noise behind mobile text. Approve the actual campaign mock-up as well as the standalone image.

4. Plan provenance, permissions and disclosure

Identify the tools and source material that will be used, and confirm the current terms that apply to those tools and inputs before work begins. A tool's ability to produce an image does not itself settle whether the team has permission to upload a reference photo, reproduce a person's likeness or use a third party's mark. Ask the responsible brand or legal team to resolve rights questions for the specific campaign and markets.

Keep a simple record of the approved references, major edits, final files and the person who signed off each version. If the workflow supports Content Credentials, they can add provenance information to an asset. The C2PA explainer is careful about the limit: provenance can describe an asset's history, but it cannot prove the image is true or factually accurate. A record complements human review; it does not replace it.

Check disclosure requirements for the exact platforms and markets in the media plan. For example, Google Ads provides an AI content label setting and notes that using the setting alone does not guarantee compliance with applicable requirements. Platform controls and rules can change, so the publishing team should verify the current position before launch rather than relying on a generic “AI-safe” promise.

5. Make approval criteria visible from the first draft

Ask the production team to show a small set of distinct directions early, not dozens of near-identical outputs. Choose a direction against the original objective: does it communicate the intended idea, maintain the brand and leave room for the required message? Only then spend time refining details. Decide what a revision round covers and when a new product, setting or campaign claim becomes a change to the brief.

At final review, check the fixed elements at 100% size and in context: product geometry, text, logo, hands or faces if present, reflections, shadows and any physical interaction. Compare with the approved source assets. Review every final crop and market version. If the image cannot meet an accuracy requirement after reasonable refinement, change the method or replace that element with a verified capture. More prompting is not always the best production answer.

A brief you can copy

Audience and intended response: [who and what they should understand]. Core visual idea: [one sentence]. Production method to test: [generation, photography, filming or hybrid]. Fixed facts: [product, brand, people and claims]. Flexible elements: [environment, light, composition and treatment]. Approved references and usage status: [links and owners]. Placements and ratios: [channel list]. Review owners: [creative, product, brand and specialist]. Disclosure and provenance plan: [who checks]. Approval test: [what must be true before release]. Final handover: [files, sizes, source assets and record of approved versions].

A strong AI visual brief gives the team room to explore while making the non-negotiables easy to verify. Visual Matters can help assess whether AI visual generation, photography or a combination fits the idea. See our selected work and explorations, or share a brief if you are deciding how to produce a campaign visual.

Let’s talkWhatsApp