Graphic design

How to design a logo

Designing a logo does not start in the software, it starts with a brief: without answers to a few specific questions about the business, every sketch is a matter of taste. The output is not an image either; it is a vector file plus a handful of versions of the same mark for the places it will actually be used.

  • Lesson 7 of 12
  • Beginner
  • Free, no signup

Six stations from a name to a usable mark

The order matters. Every station you skip is paid for by the next one.

  1. 1

    Brief

    Seven questions, one page. The smallest place of use is the important one.

  2. 2

    Research

    See the competitors side by side. The goal is difference, not ideas.

  3. 3

    Sketches

    On paper and rough. Polishing before deciding is wasted time.

  4. 4

    Vector

    Here the mark becomes a file. The small size and single colour tests happen here.

  5. 5

    Versions

    Horizontal, stacked, mark only, single colour, plus minimum clear space.

  6. 6

    Delivery

    Source file, raster exports and a usage guide. Ownership settled in advance.

This path yields a usable mark, not necessarily a good one. Taste is not built by process.

Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.

What a real brief contains

A logo brief is not a wish list of preferences; it is the set of constraints the design decisions fall out of. Seven things have to be answered in it: what the business actually does and for whom, where the logo will really be seen, which name goes in it and spelled how, what the client explicitly does not want, what other marks it will sit beside, who makes the final call, and by when.

Of those seven, one shapes the drawing more than the rest: what is the smallest place this mark will appear? An app icon and a shop sign are two different design problems. A mark built for the sign and then squeezed into a favicon usually turns into a smudge at that size.

"Something stylish" is not a brief. Our own post on the files you should receive from a logo designer says the same thing from the buyer's side: a vague brief buys you several rounds of back and forth and ends up costing more. A one page brief is cheaper than three rounds of revisions, and this is the one part of the job where buyer and designer want exactly the same thing.

The brief sheet: what needs an answer, and what arrives instead

Logo briefone page

Seven things that need an answer

  • What the business does and for whom
  • The smallest place the logo will be seen
  • The exact name and the client's preferred spelling
  • What the client explicitly does not want
  • The marks it will sit beside
  • Who makes the final call
  • Fixed dates and the schedule

What usually arrives instead of a brief

  • I want something stylish and modern
  • Like this company's, but different
  • You pick the colour
  • Send a few options and we will see

The second column holds real client answers and they are not mistakes; they are simply not a brief yet and have to be turned into questions.

What to look at before the first sketch

Research here does not mean gathering inspiration. Do not look at competitors one at a time; put them all side by side. If four competitors use a blue circle, the fifth blue circle is invisible. The goal of this stage is difference, not ideas.

Then the places the mark actually lives. Single colour printing on an invoice, embroidery on a shirt, engraving, a stamp, a favicon. Each of those is a constraint, and constraints remove options, which is exactly what you want at this stage. A design that only works on a screen has not been tested yet.

One point specific to Persian that translated tutorials never carry: a Persian logotype has to deal with letter joining and the balance of the dots. The Persian version of a mark is not the same shape with different letters; it is a separate piece of design. A mark built in Latin first and then "given" a Persian version usually breaks right here, and if you do not read Persian yourself you will not see it break.

And an honest boundary: this research is an hour or two, not a week. After that it is no longer research, it is procrastination.

From sketch to vector, and why that part is not negotiable

On paper first, rough and fast. The reason is not aesthetic: software tempts you to polish before you have decided. Twenty incomplete shapes on paper in ten minutes are worth more than two clean shapes in two hours.

Then build that shape in vector. This is the same conclusion the image formats lesson reaches from the file side: a logo is a shape, and shapes belong in vector. One vector file yields both the favicon and the three metre banner; a PNG yields neither.

Two tests with no taste in them, to be run right then: look at the mark at very small size, and make the same mark solid black. Our own post calls it the 24 pixel test, which is the size it appears at in a browser tab. A design that becomes unreadable in either state has a problem, no matter how good it looked on the monitor.

The designs that fail usually share one thing: gradients and shadows. They are what impresses in the first presentation and the first thing to disappear in real use. If a mark loses its meaning without the gradient, it is a picture, not a logo.

One logo is not enough: the versions

One file cannot fill a wide site header and also be a square app icon. That is why the output of this work is a set, not an image.

  • Horizontal and stacked lock-ups. The mark beside the name, and the mark above the name.
  • The mark alone, without the name. For social avatars, the favicon and anywhere square.
  • Single colour. Black on white and white on dark. This is the version that goes on invoices, stamps and engraving.
  • Minimum clear space and minimum size. A simple rule saying how much space stays empty around the logo and how small it may never go.

The usage guide is part of the same set, and it is the smallest part with the largest effect. Without it, six months later every employee builds their own version: one stretches the logo, one recolours it, one lays it over a photograph. An identity does not die from one bad decision. It dies from fifty small unguided ones.

Two routes, and what each actually delivers

The pans are level because each route is right for one person and wrong for another. The question is what stands behind the mark.

The fast cheap route

  • For a side project, a channel, or something you are not sure you will continue
  • You have something to put up today
  • An editable vector file usually is not part of it
  • There is no guarantee the mark is unique

The invested route

  • When the mark goes on signage, packaging and print
  • Source file, versions and a guide are delivered
  • It takes time, much of it waiting for your feedback
  • It pays back only when what stands behind it lasts

Neither pan says anything about beauty. Beauty is guaranteed neither by process nor by budget.

Delivery, and two unwelcome truths

The full list of files a client should receive, and the question of who owns the source file, belong to that same post on logo delivery and are not repeated here. For the designer the summary is one sentence: an editable vector file, raster exports at the sizes actually needed, a single colour version, and a short usage guide. Plus the question that should be settled before you start: who owns the source file after delivery.

Now the two things sales pages leave out.

A good logo does not fix a bad business. A mark is a label on something that already exists. If the service is slow, the pricing is vague or nobody answers the phone, a logo only makes that prettier and buys a few weeks. No designer will tell you this because it works against their own sale, but every designer who has worked a few years has seen it.

And from the other side, a cheap generic mark costs more later. The real cost is not the design fee; it is replacing the sign, the packaging, the app icon, the business cards and everything else printed in the meantime. Which means the main cost arrives at the point where you are no longer paying for design.

The two hold together: whether a logo is worth spending on depends on whether the thing behind it is going to last. If it is, logo and brand identity design is part of what we do too; if you are not sure yet, make that part sure first.

The fast path, with AI

The fast path here is not generating a logo with AI. It compresses the two slowest parts of the work: turning a messy client conversation into a brief, and stress testing your own mark before the client does. The image generator is the least valuable step on this path and the one everybody reaches for first. The two text steps want a frontier model rather than a cheap one, because the work is judgement; our current pick is listed in this site's AI section.

  1. Put the whole client conversation, messages and transcribed voice notes included, into the recipe below and run it. The second list in the output, the points the client never touched, is exactly what you send straight back to them. Three days of back and forth becomes one message.
  2. Only then, and only for the conversation with the client, use an image model for a mood board. Say from the start that these images are disposable: it is a way to ask "this direction or that one" with pictures instead of adjectives.
  3. Design the mark yourself, in vector. This step cannot be handed over, and the rest of the path stands on it.
  4. Now turn the model into a critic. Describe the mark in words and ask for the objections: where it breaks at small size, what it resembles, what it might read as in another culture. Models are far better at objecting than at creating, and this is the one place where their lack of tact works for you.
  5. The human pass: check Persian letter joining by eye, run the small size and single colour tests, and do the trademark search yourself at the registry. A model does none of these three.

Copy-ready recipe

Role: brand identity project manager. I am giving you the full text of a conversation with a client. Your job is extraction, not writing. Do not record anything the client did not say as something they want, and do not produce any number.

Give the brief in exactly these seven points:
1. What the business does and for whom, in the client's own words where possible.
2. The places the logo will really be seen, and the smallest of them.
3. The exact name that goes in the logo, spelled the way the client spelled it.
4. What the client explicitly does not want.
5. Competitors or marks they named, and anything they said about how those look.
6. Who makes the final call.
7. The schedule and any fixed date that was mentioned.

After the brief, give two short lists:
a. Sentences the client said that carry more than one meaning, each with the precise question I should ask them.
b. Which of the seven points above the conversation never raised at all.

If a point has no answer, write "not stated". Guessing is forbidden.

The conversation:
{conversation}

Before you trust the output: The model does not search a trademark register and cannot. Anything it says about "similar existing marks" is generated from training data, so treat it as a list to check by hand, never as clearance. And a mark that came out of an image generator is a raster picture: before anyone can use it, it has to be rebuilt in vector, and rebuilt means designed again, not auto traced over the same image.

AI in this kind of work

The honest split: AI is strong on everything around the mark and weak on the mark itself. The one thing a logo absolutely has to be is unlike the others, and a model is a machine for producing the most likely thing. So use it for the brief, for objections and for the conversation with the client, not for the design.

Tools that actually help

  • Claude For pulling the brief out of a conversation and for the critic role, because it writes out its reasoning and reasoning can be argued with. Iran is not on Anthropic's supported-countries list and there is no official payment route from Iran.
  • Gemini Its image model works for a mood board, which is the disposable imagery of step two. Google's own page says the Gemini web app runs in over 230 countries and territories, and Iran is not on that list.
  • FLUX An open weight image family, so it can be run on someone else's hardware. On Iran we know nothing: Black Forest Labs publishes no supported-countries document at all, and the empty column in our own hub means there was nothing to read, not that we did not look.
  • RGB The access and payment layer for Iran, kept separate from the tools themselves.

Where it backfires

The first risk is uniqueness, and the arithmetic is simple: similar briefs, given to the same models, produce similar marks. Thousands of other people have typed "modern minimal tech logo" and their outputs are from one family. The second is ownership, and here we deliberately make no legal claim: whether a mark can be registered depends on where you register and on what is already on the register, the generator does not answer that and neither do we; the search at the registry is the buyer's job and doing it in the first week costs nothing. Third, traces: Google writes that the SynthID watermark is embedded inside the image at the moment of creation and is designed to survive cropping, filters and lossy compression, so converting the format does not remove it. And fourth, one that is ours specifically: image models draw Persian letters as shapes that look like letters. The joins break and the dots land wrong, and if you do not read Persian you will not see it, but your client will.

Sources: Google DeepMind: SynthID Google Search Central: favicon requirements for Search

Where this advice stops

This path is for a mark that has to last. If you are testing a side project that may not exist in three months, the whole process is over engineering, and a quick mark plus rebuilding it later is the cheaper bet; our own post on logo delivery says the same about the three routes. The second limit matters more: this process does not produce taste. What it delivers is a mark that survives its uses, which is a lower and far more checkable bar than a good logo. The good designer is still the one who, after all of this, draws the better shape.

From our own work

The brand mark on this very site is not an image at all. On 2026-09-06 the rgb.ir homepage carries zero img tags, and what you see as the logo is text with a CSS gradient, so it is sharp at any size and no file is downloaded for it. The only place this same logo was forced to become a raster file is where a platform demands fixed sizes: the site icon set, three WebP files at 32, 180 and 192 pixels cut from one master, weighing 1146, 9002 and 9404 bytes. Exactly the pattern this lesson describes: one master, a few exports, only where they are needed. And a mismatch we found on our own site while writing this lesson, which honestly is still not fixed: Google's own documentation lists the favicon formats Search supports as BMP, GIF, ICO, PNG, JPEG, PPM and TIFF, and WebP is not among them. We do not know what Google actually does with our icon; we know the list does not include the format we used.

Real follow-up questions

How long does designing a logo take?

In our own process, seven to fourteen working days, and three to six weeks for a full identity; those same numbers are published on our graphic design services page. What stretches that range is not drawing speed but how fast feedback comes back. Anyone selling "a logo in 24 hours" has removed one of the stages, and it is usually the brief.

Should I make my logo with AI?

For a side project, or something you are not sure you will continue, yes, and there is no reason to spend. For a business that will print signage and packaging, no; and the reason is not aesthetics but two other things: an editable vector file that is genuinely yours, and some confidence the same mark was not handed to ten other people.

What is the difference between a logo and a brand identity?

A logo is one element; a brand identity is the system that element lives in: the colours, the type, the layout rules and the usage guide. A small business usually needs the logo and its versions first, and builds the rest of the system later and gradually.

Keep learning