You see a striking image, a polished video or an impressive website online. Beside it is a short explanation: “Made with AI.” Sometimes the prompt is shared too.
If you already have a direction you like, it can be a starting point. The review should show what can be kept and what preparation it still needs for its intended use.
It is natural to assume the whole result came down to that instruction. In practice, a result can come from a single suitable generation or from a process with references, selection, revision and preparation for use.
The number of attempts alone does not describe the value of the work. What matters is what needs to be achieved and how we check that the result meets that requirement.
The prompt starts before it is written
For a campaign, we need to know the product, the audience, the message, the environment and the desired action. We need to identify what can be explored freely and what exactly must be kept: the product’s shape, the color, the number of elements, the copy or the terms of the offer.
“Make it look premium” leaves many decisions open. If the campaign needs room for a headline and a product that remains recognizable on a phone, spell those requirements out and check the result against them.
Google’s prompt design documentation treats clear instructions, context and experimentation as part of the process. That does not guarantee a perfect result on the first attempt. Google: prompt design strategies
A demonstration made for this article
For this demonstration, we used a fictional AI-generated bag. It is not a photograph of a real product or a commercial portfolio project. The three images show an initial generation followed by two revisions.
The first brief asked for a campaign image of a green bag with brown handles, in a weekend setting. The result was rich in context: the bag, vehicle, plants and landscape suggested a trip away.
That image could work for an atmospheric introduction. For a website banner with a headline on the left, however, too many background elements competed for attention. The next decision followed its intended use: simplify the scene and leave room for the message.
Give each revision a purpose
In the second request, we kept the fictional product’s direction and asked for a simple, warm background with the bag on the right. The vehicle, animal, plants and other contextual objects were removed.
The change creates space for copy and clarifies the hierarchy. It does not prove that the image will sell more. That would need assessing in the context of the campaign and audience.
For the third request, we adapted the composition to portrait format, leaving space above the product. Cropping the landscape image would not automatically produce that arrangement.
| Step | Decision | What we check |
|---|---|---|
| Explore | Build a weekend atmosphere | Scene relevance and product clarity |
| Composition | Simplify and leave space for copy | Hierarchy and message placement |
| Adaptation | Reorganize for vertical format | Framing and legibility at actual size |
Space for copy depends on the intended use
We asked for room for a message because the example is intended for a website composition or a layout controlled by a designer. The copy can remain a separate HTML element, keeping it readable and adaptable on mobile.
A separate image asset in a platform-assembled ad may need a different brief: a larger product in the frame, no embedded headline and framing suited to the specification. A website composition should not automatically carry over to every campaign. See Size matters. So does context.
Decide how the image will be used before writing the prompt. It may change what you ask for and avoid revisions discovered only when the asset is uploaded.
What can slip through even after a good review
A model can subtly alter stitching, proportions, lettering or accessories. Our fictional bag retains a recognizable direction across these images, but the sequence is not proof of exact reproduction of one object.
For a real product, compare the output with its photographs and specifications. If the required fidelity cannot be achieved, choose another method, such as product photography with carefully limited changes to the surrounding scene.
We also check copy, claims, areas covered by the placement’s interface and the final file. An attractive image may still require layout preparation and implementation.
When we stop
We stop revising when the agreed criteria are met and another change would bring no justified benefit. Sometimes, the first version is enough. Sometimes, a precise change resolves the problem faster than a series of general requests.
More iterations do not automatically mean more quality. A good brief, proper references and verification can reduce unnecessary attempts.
At NOT MORE., we want to use AI in a process where each intervention has an explanation. We choose, check and prepare the result for where it will be used.
How we present the process so it can be evaluated
When we show a project made with AI, the selection of images and the explanations around them become part of the demonstration. Readers need to be able to tell apart the direction explored, the adaptation decision and the result prepared for a specific use.
In this article’s bag example, we can demonstrate the change in composition and in the space available for copy. We cannot demonstrate a campaign’s performance, because none was measured here. Nor can we present the images as a verified reproduction of a real product: the object is fictional. These clarifications let readers evaluate exactly what the example shows.
For commercial work, a useful presentation explains the brief, the important constraints, each team’s contribution and the criteria by which the delivery was accepted. If we show a before-and-after comparison, we say what we changed and why. Differences in format, framing or use have to be visible, so we do not compare results built for different tasks.
What a case study can demonstrate
A story about how the material was made can explain the reasoning. An implemented file can demonstrate the delivery. A report can describe measured results. These kinds of evidence answer different questions and are worth presenting as such.
If we include commercial data, we state the period, the source and the relevant changes in the project. A shift in sales after the material was published does not, on its own, establish the effect of the image or the AI tool. The offer, the budget, the distribution and the buying process can all contribute to the result. Without an evaluation that separates these contributions, we state the conclusion within the limits of the information available.
Likewise, claiming time savings requires an explained comparison: between which processes, for which deliverables and with which checks included? Generating may take very little time, while selecting, correcting, laying out and preparing the files may make up the rest of the work. We do not attribute the duration of a single operation to the whole project.
At NOT MORE., we can show both the result and the decisions relevant to using it, while respecting the agreed confidentiality and rights. The purpose of the presentation is to make the work possible to evaluate and to explain under what conditions it could be useful in another project.
Three questions about a generated result
- What use was this designed for, and which conditions must it meet?
- What should be kept exactly and how has it been checked?
- Is the material ready to publish, or does it still need adaptation and implementation?
What does your result still need before it can be used?
Show us the current version, what you want to keep and where it will appear. We’ll discuss the criteria for checking it and the changes that would have a clear purpose.