DIY: make your own brand identity with AI — logo, banner, cover template and a style rule, in an afternoon

You have a channel, a small product, a service, a game — and it needs to look like something before anyone sees it. The first search result is a marketplace of logo designers; the second is a logo generator that gives everyone the same swoosh. This is the third option: the recipe I use for every new project, including the seven video channels in my channels log — each with its own identity, none of which cost more than a coffee.

You end up with

A name you've tested, a written style rule, a logo/avatar, a banner, a cover template, and a brand folder every future asset gets checked against

What it costs

Cents: image generations at roughly $0.03 each on a pay-per-image API — a full identity is 10–30 generations. Chat assistant: free tier or an existing subscription

How long

One afternoon. The doctrine takes 20 minutes and saves hours later

Runs on

A chat assistant · an image model (API or subscription) · a free vector tracer if you need SVG · a folder

Step 1 — test the name before you draw anything

Ask the chat assistant for twenty name candidates with the constraint that matters to you (short, says what it does, works in your languages, readable in a video thumbnail). Then do the boring part by hand: search each survivor as a domain, as a handle on the platform you’ll live on, and in your country’s trademark register and the EU one. Ten minutes per name. A name that is taken on the platform you need is not a name.

Keep the one that survives and that you can say out loud without explaining. Wordplay that needs a footnote dies in a thumbnail.

Step 2 — write the visual doctrine (this is the actual product)

One paragraph, written before a single image exists. It states what the brand looks like and what it never looks like. Mine for one of the channels, verbatim from the brand folder:

Full-length caricature: big head, recognisable features, period clothing and props. Strictly not a children’s-cartoon look. One style for every character. Almost static, with a little motion.

And for another:

One static courtyard at dusk, only the streetlamp flickers, digitised-film grain, no people in frame.

The doctrine is what makes the next hundred assets look like they belong together. Without it you’ll generate ten gorgeous images in ten styles and have no brand. With it, every prompt starts from the same sentence.

Step 3 — the logo and avatar

Prompt the image model with the doctrine plus the job: “avatar for a channel about X, reads at 100 px, centred single subject, flat background, no text”. Generate 6–10 with a fixed seed where the model allows it — a fixed seed means small prompt edits change the image slightly instead of completely, which is how you converge instead of gambling.

Three rules that save regenerations:

• No text in the image. Models still misspell. Put the wordmark on afterwards in any editor.
• One subject, centred, plain background. Avatars are tiny; detail is noise.
• Ask for the shape you’ll crop to. Square for avatars, 16:9 for banners, 9:16 for story covers — don’t crop a landscape into a square and lose the subject.

If you need a vector (print, laser, embroidery), run the chosen PNG through a free tracer and clean it in a vector editor. For screens, the PNG is enough.

Step 4 — the banner and the cover template

The banner reuses the doctrine at 16:9 with room on one side for the wordmark. Generate it from the same seed and prompt family as the avatar so the two visibly belong together.

The cover template is the asset most people forget and the one you’ll use most: a fixed layout — where the image goes, where the two to four words of text go, which colour the text is — saved as a file you copy for every new video, product or post. For the channels I keep a one-line caption rule (“warm, one line, over the whole frame”) next to the layout so it never has to be decided again.

Step 5 — the brand folder

Everything above goes in one folder: doctrine.md, the seed and the prompts that produced the winners, the avatar, the banner, the cover template, the palette as hex values, the font names. Every future asset is checked against this folder, and when a future version of you — or a designer you finally hire — asks “what does this brand look like?”, the answer is a folder, not a memory.

Where it breaks

• The identity is generic. Almost always because the doctrine was generic. “Modern, clean, minimal” produces the same image for everyone. Add a constraint that is yours: an object, a time of day, a medium (film grain, woodcut, chalk), a prohibited thing.
• Faces. Image models render faces well and differently every time, which kills consistency. For a recurring character either lock the seed and crop tightly, or design the character so the face isn’t the identifier — from behind, hooded, silhouette, far away. The channels use the second approach; it also sidesteps the question of whose face that is.
• Style theft. “In the style of [living artist]” is a bad habit and, increasingly, a policy violation. Describe the medium and the mood, not a person.
• The name was free everywhere except the one place you needed. Step 1 exists because I’ve skipped it.

What it gives your business

A brand that exists on day one instead of day thirty, for the price of a coffee, with a brief you own. More importantly: you now know that the thing agencies charge four figures for is a doctrine plus twenty generations plus taste — and taste is the part you were always going to supply.

Related: personal brand for freelancers · the tools I actually use · all recipes in Do it yourself.