Start from a brief, not a picture
The usual approach to cover art is to wait for an image to arrive, then try to make it. That is why so many single covers are a photograph of the artist looking away from the camera: it is the idea that shows up when no other idea does.
Working designers start somewhere else. They write a brief first: a few sentences describing what the image has to be, in terms concrete enough to argue about. “A cold, almost empty image, blue and grey, with one small human figure near the bottom edge” is a brief. “Something moody that fits the song” is a wish.
The difference matters because a brief can be executed several ways and evaluated honestly, while a wish can only be satisfied by luck. It also means you can hire someone, or use a stock image, or shoot something yourself, and still end up with the same cover.
Everything below is how to write that brief from the song you already have, and then how to make sure the result survives contact with a streaming app.
Reading the imagery out of the song
The lyric is the raw material, and the mistake almost everyone makes is going after the wrong part of it.
Take the concrete nouns. Leave the metaphors alone. If a line mentions a kitchen table, a coat, a bus window or a set of keys, those are usable: they are objects that can exist in a photograph. If a line says the narrator is drowning, or burning, or falling apart, do not photograph someone underwater. The metaphor is already doing its work in the words, and illustrating it literally is the single clearest signal of an amateur cover.
The images that work sit beside the lyric rather than repeating it. A song about a relationship ending, whose cover shows an unmade bed and a single coffee cup, is doing something. The same song with two silhouettes walking apart is doing nothing the title did not already do.
- List every physical object named in the lyric. Most songs contain between three and eight.
- Note the setting. Interior or exterior, day or night, city or nowhere.
- Note who is present. One person, two, a crowd, nobody.
- Ignore the adjectives. They describe feeling, and feeling is carried by colour and light rather than by subject.
Concrete action: pick one object from your list and build the image around it. Covers that contain one thing are stronger than covers that contain a scene, and they are dramatically stronger at small sizes.
Mood and colour
Colour does more emotional work than subject matter and is the cheapest thing to control. It is also where most self-made covers go wrong, and the failure is nearly always the same: too many colours.
Two colours plus a neutral reads as designed. Five colours reads as accidental, because it is the palette you get when you do not choose one. This holds regardless of genre and regardless of whether the image is a photograph or an illustration.
- Warm and saturated: intimacy, memory, physical closeness.
- Cool and desaturated: distance, control, isolation.
- High contrast with few colours: confidence, aggression, immediacy.
- Low contrast and soft: dream, ambiguity, nostalgia.
Match the palette to the production rather than to the genre. A warm, roomy, imperfect recording sits against warm imperfect colour; a cold, precise, heavily processed track sits against controlled colour. When they disagree the cover feels wrong for reasons listeners cannot articulate, which is worse than being plainly bad.
Concrete action: choose the two colours before you choose the image. It constrains every later decision usefully, including wardrobe and location if you are shooting.
Face, object or abstraction
Nearly every cover in existence is one of three treatments, and picking one on purpose is most of the design work.
The face. A person, usually the artist. It builds recognition across a catalogue, which is why established artists use it constantly. It is also the hardest to execute, because a mediocre portrait reads as a mediocre artist, and it requires a photograph good enough to survive being two centimetres wide.
The object. One thing, isolated, well lit. The most reliable option for an independent release: cheap, distinctive, memorable at any size, and it does not depend on anyone being photogenic. If you are unsure, this is the correct default.
The abstraction. Texture, colour field, shape, type alone. Forgiving and contemporary, and it carries a risk: abstraction without a strong idea becomes generic instantly, and generic abstraction is invisible in a grid of thumbnails.
Concrete action: if this is an early release and you are building recognition, alternate. Face covers for the singles you want associated with you, object covers for everything else. Consistency across a catalogue is worth more than any single cover.
Typography is part of the concept
Type is not something applied after the image is finished. It is a component of the concept, and treating it as an afterthought is why so many otherwise decent covers look unresolved.
- One typeface. Two weights of it if you need hierarchy. Two different typefaces on a square this small almost never works.
- Decide early whether text appears at all. Plenty of strong covers carry no type, since the platform displays the title beside the image anyway.
- If it appears, make it large enough to survive. Small elegant type at full size becomes an illegible smudge in a playlist row.
- Give it room. Type crammed into a corner reads as a caption; type placed deliberately reads as design.
- Avoid the default fonts that ship with design apps. They are recognisable and they signal the amount of thought that went in.
Concrete action: produce one version with type and one without, then compare both at thumbnail size before deciding. The answer is frequently the version without.
The fifty pixel test
This is the step that separates covers that work from covers that only work on your monitor, and almost nobody does it.
Your artwork will be seen at full size a handful of times and at 50 to 100 pixels thousands of times: in a playlist row, in a search result, on a lock screen, in a friend’s shared story. The small version is the real cover. The large version is a detail view that few people ever open.
- Export at 3000 pixels, then view it at 50. Shrink the window until the image is the size of a fingernail.
- Ask whether anything is still readable. Not the type, the image. Is there a recognisable shape? A colour you would recognise again?
- Check it against real competition. Open a playlist in your genre and put your cover in that grid mentally. Does it disappear?
- Test on a phone, not a laptop. Different screen, different brightness, different size.
What survives at that size is contrast and simplicity. What dies is fine detail, subtle gradients, small type and busy photographs. This is precisely why the isolated object beats the complicated scene, and why two colours beat five.
Concrete action: if the cover is unrecognisable at 50 pixels, do not adjust it. Redesign it around one shape and one colour relationship, because no amount of tweaking rescues an image that is fundamentally too busy.
Delivery specs and what gets rejected
Distributors reject artwork for a small, predictable set of reasons, and each rejection costs days on a release timeline. None of them is difficult to avoid.
- Size and shape: 3000 by 3000 pixels, perfectly square, RGB colour, JPEG or PNG. Anything smaller than 1400 is generally refused outright.
- No URLs, no social handles, no logos of streaming services or stores. This is the most common cause of rejection.
- No promotional text. “Out now”, “New single”, prices, dates and barcodes are all disallowed.
- Text must match the metadata. If the cover says a title or artist name, it has to be the same string you delivered, character for character.
- You must own or license everything in it, including fonts, stock images and any photograph of a person, who has to have agreed.
- No blurred, pixelated or low-resolution imagery. Upscaling a small file does not pass.
Concrete action: keep the layered source file. You will need a version without type for social, a vertical crop for stories, and eventually a version at a different size for something you have not thought of yet. Recreating a flattened cover a year later is miserable.
Make three, not one
The final piece of advice is procedural rather than aesthetic, and it changes outcomes more than any single design decision.
Do not make one cover. Make three genuinely different concepts: not three variations of the same image with the type moved, but three separate answers to the brief. One face, one object, one abstraction, or three different objects with different palettes.
The reason is that the first idea is almost never the best one, it is simply the first. Having three forces the comparison that reveals what you actually want, and it makes feedback useful: asking someone “does this work?” produces politeness, while asking “which of these three?” produces an answer.
Then test all three at fifty pixels, in a grid, on a phone. The one that survives is frequently not the one you were attached to, which is exactly the point of doing it before the release rather than after.
Deriving those three concepts from the song rather than from a mood board is what keeps them coherent with the music. The song has already made most of these decisions: its themes, its imagery, its emotional register and its intended audience are all in there, and reading them out is cheaper and more honest than borrowing a look you admired on someone else’s release.