Logos, and Why They Are the Hard Case

The first thing everybody asks for

Almost nobody opens an image tool for the first time and asks for a background texture. They ask for a logo. It is the most obviously valuable design object a small business owns, the quote from a studio can be eye watering, and here is a machine offering to produce one in fifteen seconds.

This is the lesson where the course tells you not to. Not because generated images are bad, and not out of loyalty to designers, but because a logo is a genuinely unusual object with requirements that image generation is structurally poor at meeting. Understanding why is more useful than being told no, because it also tells you exactly which parts of logo work AI is excellent for.

What a logo actually has to do

A logo is not a picture of your company. It is a piece of engineering with four jobs.

It has to work at one centimetre and on a billboard. The same mark, unchanged, has to be legible as a tiny circle on a phone screen and as a two metre shape on a van. That forces simplicity. Anything with fine gradients, delicate line work, small internal detail or a bit of texture will turn to mush at small sizes, and generated logos are almost always over decorated because detail is what these tools are rewarded for producing.

It has to work in one colour. Sooner or later your mark will be embroidered on a polo shirt, etched into glass, stamped on a receipt, faxed by a supplier who somehow still faxes, or printed in black on a form. If the design only reads because of a colour transition, it stops existing in those places.

It has to be reproducible in every format. Real logo work ships as vector artwork, because vector stays sharp at any size and can be handed to a sign maker, an embroiderer or a printer without argument. Image generators produce raster images, made of pixels. A pixel image can sometimes be traced into vector afterwards, but tracing a fussy generated mark tends to produce a technical mess rather than clean shapes.

It has to be defensible. This is the part people discover last and hate most. A logo is only worth something if it is distinctive, if you can keep using it, and if nobody can plausibly claim you took theirs. That is not a design question. It is a research question, and it happens before anyone draws anything.

Why generated logos disappoint

Put those four jobs next to what you saw in lesson one and the mismatch is obvious.

Generated marks tend to be busy, because busy looks impressive in a preview and the tool has no idea it will end up 8mm wide on a business card. They frequently include text, which is the classic failure point, and a logo with a subtly wrong letterform is worse than no logo. They arrive as pixels rather than shapes. And most quietly of all, they echo things the system has seen a great deal of, which means the swoosh, the leaf, the abstract hexagon and the geometric animal head that half your industry already uses. You can end up with something that feels familiar in a bad way, and familiar is the exact opposite of what a mark is for.

There is also a practical trap. A generated logo often looks great in the square image it arrived in and falls apart the moment you put it on a real thing, because it was never tested against the actual jobs. If you take nothing else from this lesson, take the test: shrink it to the size of a fingernail, turn it to solid black, and put it on a photograph of a shirt. Most generated marks fail all three.

What AI is genuinely brilliant at in logo work

Now the useful half, because there is a great deal of it.

Exploration. Before anyone commits, you need to know what direction feels right. Generate forty rough marks in half an hour, spread them out, and notice what you keep returning to. None of these will be your logo. All of them are helping you find out that you want something geometric and quiet rather than illustrative and warm, which is information you did not have this morning.

Vocabulary. Most people cannot say what they want, which is why briefing a designer is hard. Working through generated options gives you the words. "Not that, it looks like a law firm" is a genuinely useful sentence, and it usually only arrives once you have seen the thing.

Testing a concept before you pay for it. If your idea is a lighthouse, you can find out cheaply whether a lighthouse looks like every other lighthouse in your sector, or whether it can be reduced to something simple enough to survive being tiny.

Everything around the logo. Once you have a mark, the supporting world of a brand is much friendlier territory: background textures, illustration for a website, social imagery, patterns, spot drawings. Lessons four and six are largely about that, and it is where most of your ongoing value will actually come from.

Writing the brief for a human. This is possibly the highest value use in the whole lesson. Use a chat tool to turn everything you learned during exploration into a proper written brief, then hand that to a designer. Better brief, less back and forth, and a noticeably better result for the same fee.

I run [what the business does, in one sentence, for whom, where].
I have been exploring logo directions and here is what I learned about my own taste:
I keep being drawn to [describe it]. I keep rejecting [describe it] because [why].
Write me a logo brief for a professional designer. Include:

What the business is and who it serves.
The feeling the mark should give, and the feelings to avoid.
Where it has to work: smallest and largest use, single colour uses, physical items.
What I am NOT asking for, so nobody wastes time.
The questions the designer should ask me that I have not thought of.
Keep it to one page. Plain language, no design jargon I would have to look up.

When to pay a human

Not always. Be honest about which of these you are.

If you are testing an idea, running a side project, launching a market stall, or genuinely do not know whether the business will exist in a year, a placeholder mark is a completely reasonable decision. Set the wordmark in a clean typeface, keep it simple, get on with the actual work. Plenty of businesses run for years on a tidy piece of type and lose nothing by it.

If the logo will go on signage, packaging, vehicles or a shopfront, if you intend to register it, if you are raising money, or if the cost of changing it later would be painful, that is when you pay somebody. Not because AI cannot make something attractive, but because you are buying research, options, a proper set of files, and a person who is accountable for whether it works on a polo shirt. That is a different purchase from a picture.