An AI Logo or a Designer: Where the Real Line Runs

It is framed as a choice, and it is not one

"AI or a designer" sounds like competition. In practice they are two stages of the same job.

Generation is strong where volume helps: an hour shows more directions than a person draws in a day. It is weak where consequences have to be owned: print, reduction, rights, resemblance.

So the productive question is different: which part of the work can be handed to generation, and which part cannot.

What generation does well

It widens the search field by an order of magnitude.

It removes the main constraint of the early stage — exhaustion with your own ideas. Twenty directions in an hour let you see the whole territory rather than the first three obvious answers.

That also makes it a useful conversation tool: a client finds it easier to point at an image than to describe one.

Where it fails consistently

Three places, and all three surface afterwards.

First, format. The output is raster: a fixed pixel count that falls apart on a shopfront sign. Redrawing it in vector is done by a person.

Second, verification. A generator does not shrink the mark to an icon, print it in one colour, or set it on a dark ground. Third, resemblance: it consults no register and does not know what it has echoed.

Rights

This is the most underrated item on the list.

Terms for using a generated image differ between services and jurisdictions, and there is no universal answer. They need establishing before the mark goes onto packaging.

What matters here is not a position for or against generation, but that the question gets asked in writing at all.

The division of labour that works

Generation on the search. A person on the decision, the verification and the assembly.

In practice: a fast sweep of directions, then one choice, then manual vector drawing, then verification in black and white, at small size and on real grounds, then rules of use.

Speed survives this: the bottleneck in branding is almost never the drawing.

What it means for a budget

Generation does not remove the cost of verification, because verification is the bulk of the work.

It shortens the search, and that is a real saving. But a mark that cannot be printed and whose rights are unestablished has not become cheaper — it has deferred the invoice.

An honest estimate counts both halves: the fast search and the mandatory finishing.

What generation looks like in practice

Discussion of these tools tends to stay abstract, so it helps to describe how it looks inside a real project.

The first hour: the task and its constraints are written down, then the sweep begins. In that time several dozen directions can be reviewed — not finished marks but images, moods, families of form. Some are discarded at once, some set aside.

The second stage is narrowing. From what survives, the candidates that could actually become a mark are separated out: simple forms that work in one colour, with no fine detail. Most of the striking results die here, because they are good as illustration and unusable as a mark.

The third stage is a person. The chosen direction is drawn again in vector: curves built, proportions aligned, spacing set. This is no longer tracing a generated image but constructing a form on its theme.

Only then do the checks begin: reduction, one colour, dark ground, a real photograph. The time saved landed in the first stage; generation shortened none of the ones after it.

Why automatic tracing does not solve it

The temptation is obvious: the output is raster and vector is needed, so run it through automatic tracing. The tool exists in every vector editor and takes a second.

The problem is what comes out. Tracing places nodes where the pixel colour changes, not where the form changes. A simple arc ends up with thirty nodes instead of three, the line wobbles slightly, and corners round off unevenly.

On screen at a large size the difference is barely visible. It appears in print, in vinyl cutting for signage and in engraving — that is, in exactly the places vector was needed for.

There is a second problem: such a mark cannot be edited cleanly. Moving an element half a millimetre means rebuilding thirty nodes by hand, and the next supplier usually prefers to draw it again.

So tracing is acceptable as a rough stage and useless as a way to obtain a source file.

What generation changes in cost and time

The expectation that generation collapses the price does not survive contact with practice, and the reason is arithmetic.

Of the four line items in a quote — framing, directions, verification, assembling files and rules — generation accelerates exactly one. The other three remain human work and take most of the time.

The real effect is different and also valuable: the search field widens. The same hour covers more options, and the chance of finding something other than the obvious rises. That affects quality rather than price.

The second effect is communication. A client finds it easier to point at an image than to describe one, so agreement on direction arrives sooner. That shortens calendar time, which as we covered depends more on decisions than on production.

So the honest answer to "has it become cheaper" is: faster during the search, unchanged everywhere else.

Where generation belongs beyond the mark

Outside the logo the territory is wider and calmer, because the format requirements are softer.

Backgrounds and textures: what used to be searched for in stock or shot on purpose can now be generated for a specific task in the required palette. Here raster is the norm rather than a limitation.

Filling layouts: instead of grey rectangles, a deck or a site mock-up carries meaningful images. The conversation becomes concrete immediately, because everyone can see how the layout behaves with a picture rather than a placeholder.

Mood boards and the search for a visual language: here speed of iteration converts directly into the quality of the conversation with a client.

What still should not be done: substituting a generated image for a photograph of a real product or a real room. That is not a question of technology but of honesty toward a buyer.

What to ask if a supplier uses generation

Using generative tools is in itself neither good nor bad. What matters is what happens afterwards.

First question: will the final mark be drawn in vector by hand, or will it be a trace? The answer decides whether the mark has to be redrawn in six months.

Second: who checks the result against existing marks for similarity, and at what stage? A generator does not, and if nobody else does, the risk is merely deferred.

Third: what are the terms of the service the result came from, and how is that reflected in the contract? An answer of "it does not matter" means the question was never studied.

Fourth: what will the handover contain? Here the answer should not differ from any other project, and that is a good sign: the file set does not depend on what was used to find the direction.

It is also worth adding that tools in this area change faster than practices around them settle. So a contract is better written about requirements for the result — vector output, similarity verified, rights established — than about a particular service. Wording like that survives a change of tool, whereas naming one does not. The same holds for internal guidelines: describe the standard the work has to meet, not the software somebody happened to use to reach it.