Brand guidelines that survive AI production

A production process that makes two hundred images a week has none of those things. It cannot interpret and it will not ask. It can only follow what is specific, and guess at what is not. Guessing at volume is what drift looks like.

This article is about the gap between a guideline a designer can use and a guideline a production process can use, and what to add before you start generating.

Three kinds of instruction that stop working

Instructions that describe a feeling. "Warm and inviting", "premium but approachable", "clean, Scandinavian". These are useful for alignment in a meeting and useless as production instructions. Two competent people will read them differently, and two generations will differ more than that.

Instructions that assume a shared past. "Like the autumn campaign." Everyone who was there knows what that means. Nobody who joins next year does, and neither does any process.

Instructions with a hidden exception list. "Always on white, except for the outdoor range, and except when it is for the catalogue, and the new line has its own thing." The exceptions live in one person's head. When they are on holiday, the exceptions become the rule or disappear entirely.

If your guidelines are mostly these three, you do not have a brand problem yet. You will have one the first week you produce at volume, because volume turns small ambiguities into a visible pattern.

What to add before you generate

Six additions. None of them requires a rewrite of your brand book, and all of them can sit as an appendix to what you already have.

1. Numbers where you currently have adjectives. Background value, not "white". Light direction as a position, not "soft". The angle set, in order. Framing as a margin. Aspect ratios as a list. A designer does not need this; a process does.

2. Reference images, one per rule. This is the highest-value thing on the list. One approved image next to each rule beats a paragraph describing it, because everyone who produces for you can hold their result next to your reference and see the answer immediately. A guideline you can compare against is enforceable. A guideline you have to interpret is a conversation.

3. Negative rules. What must never appear. Competing products in the background, visible brand names that are not yours, a season that contradicts the launch, a colour you do not sell, people whose faces are recognisable. Positive rules describe your ideal; negative rules catch the plausible mistakes, and at volume the plausible mistakes are the dangerous ones. Nobody misses a disaster. Everybody misses a kitchen tap in a colour that is not in your range.

4. Material notes per product family. Brushed metal, glass, woven fabric, transparent plastic and anything with a fine repeating pattern are where generated images give themselves away. Write down, per family, what "correct" looks like and what the usual failure is. This is the single most useful page you can add, and it is the one page nobody has.

5. Output definitions per channel and per market. Which images, in which formats, for which channels and countries. Write the list once; it is the same list every time. Without it, every launch re-decides what "done" means, and the answer gets smaller every quarter.

6. An escalation rule. What happens when something falls outside the guidelines. Who decides, how fast, and where the decision is written down afterwards. Every exception either joins the reference set or is refused. Guidelines without this become an archive of what you used to think.

What not to add

There is a failure mode in the other direction, and we have walked into it ourselves. Over-specified guidelines produce images that are technically compliant and completely lifeless, and they make every new product a rules dispute.

Two things help. Keep the fixed core small, five or six decisions that make your product recognisable. Mark everything else explicitly as flexible, so people know they are allowed to vary it rather than discovering it the hard way.

The point of a guideline is not to remove judgement. It is to spend judgement on the things that matter instead of on the background colour, again.

The part that is not a document

A guideline only works if something checks against it. At one image a day, the check is informal: somebody sees everything. At two hundred a week, informal means nothing gets checked.

So decide three things explicitly:

What gets checked. Everything that leaves the building, or a defined sample plus everything in a risk category.

Who checks. A named person, not a team in general.

What happens to a rejection. A rejected image that teaches nobody anything is a cost. A rejected image that becomes a negative rule is an investment.

This is why we keep a person from our team in the loop on everything we deliver. Not as a philosophy. The failure mode of generated imagery is not a spectacular mistake, it is a plausible one, and plausible mistakes are exactly what an unattended process ships.

While you are at it, record where each image came from: generated or photographed, which tool, who approved it, for which markets it is cleared. We wrote about why that field matters in recording where an image came from.

A one-page test

Take your current guidelines and hand them to somebody who has never worked on your brand, with one product and no further explanation. Ask them to describe exactly what the image should look like.

Everything they get wrong is a gap. Everything they have to ask about is a gap you were paying for in meetings. Fix those, and stop there; the rest of your brand book is probably fine.

We wrote about what belongs in the image chapter specifically in what belongs in the image chapter, and about where consistency breaks as a catalogue grows in where consistency breaks.

Where LensLab fits

We work from a Brand DNA per client: the rules above, captured once, so production stays inside your own rules instead of drifting toward whatever the model finds attractive. Our creators produce the visuals, a person checks them before delivery, and you only pay for the visuals you accept. You can see how it works in more detail.

One thing we are building rather than selling: a Brand Book Builder is coming, which will make capturing these rules part of the platform. It is not available yet, so today we set this up with you.

In closing

AI did not break brand guidelines. It read them accurately, which is a different and more embarrassing problem. Make the rules specific, add reference images and negative rules, define what "complete" means per product, and put a person on approval.

Want to see what that looks like on your own range? Try LensLab free with five credits.