AI cover art that does not look like AI cover art
Whether you should use AI art is a question about your book, your readers and your conscience. This guide is about the other question: if you are going to use it, how do you keep it from announcing itself?
Because it does announce itself, and readers have gotten fast at spotting it. Almost every tell comes from the same habit — treating the generated image as a finished cover instead of as raw material.
The tells
- The sheen. Default output has a glossy, over-rendered look where everything is equally lit, equally sharp and equally important. Real cover art has a focal point and lets the rest fall away. That uniform polish is the single most recognizable signature.
- Faces and hands. Hands are still the classic failure. Faces are subtler: unnaturally symmetrical, skin with no pores, eyes that are slightly too matched.
- Text baked into the image. Generators love to invent book furniture — a fake review quote across the top, a made-up imprint at the bottom, the words "A NOVEL" in a typeface you did not choose.
- No quiet area. The composition fills every square inch and puts the subject dead center, leaving nowhere for a title that isn't fighting the art.
- Lighting from nowhere. Rim-lit figure, volumetric god rays, teal and orange haze. It reads as generic fantasy regardless of what you asked for, which means it signals nothing to a reader scanning a category page.
Design, do not illustrate
The instinct is to describe the scene: the lighthouse, the storm, the woman on the cliff, the locket. That produces an illustration of a scene, and a cover is not one.
Ask for a composition, a palette and a mood instead. One subject, a stated color range, a named quality of light, and an explicit instruction about where the frame should stay empty. You will get something plainer and far more usable, because a cover's job is to be read across a room, not examined.
If you cannot say what the one subject is, the generator cannot either, and you will get all of them.
Leave the space before you generate
Type dropped onto finished artwork is an amateur tell that predates AI by decades, and generated art makes it worse.
Decide where the title goes first — upper third, lower third, a band down one side — and ask for that area to stay quiet: open sky, shadow, water, wall, fog. Then generate ten variations and keep the ones with room, not the ones with the best subject.
If the art is finished and there is nowhere for the title, generate again. Rescuing it with an outline and a drop shadow never works.
In the Cover Maker. Title, author and series lines are live text layers kept as vector in the exported PDF, so your type is never baked into the artwork — move it, resize it and recolor it against the image as often as you like. Start a book → · about the Cover Maker
Generate at the shape of your book
Generating a square and cropping it later throws away half your pixels and usually the composition with them. Start at a ratio close to your trim:
- 2:3 suits 6 × 9, 5.5 × 8.5 and 5.25 × 8.
- 10:16 suits 5 × 8.
Bleed adds a little height to the bare trim, so expect to crop slightly — but from a ratio that is already close, not from a square.
Upscale before you download
This is the part most authors miss, and it is the one that shows in print.
A typical generated image is about 1024 pixels on the short edge. On a 6 × 9 cover that works out to roughly 170 DPI — well under the 300 DPI print target, and firmly in the range where a printed cover looks soft rather than sharp. It will look perfect on your screen the entire time.
So use the generator's own upscale step before you download, every time. Enlarging afterward is repair work; generating large is free.
In the Cover Maker. The tool checks your front artwork against the print guideline, flags anything under 200 DPI as visibly soft, and offers one-click enlargement to 300 DPI. That enlargement runs inside your browser — ordinary high-quality resampling, no AI, and nothing uploaded anywhere. Start a book → · about the Cover Maker
Treat it as raw art
The generated file is your starting material, not your cover. What turns it into one:
- Crop. Reframe for the title. Off-center is almost always better than the composition you were handed.
- Color-correct. Pull the saturation down, push the contrast where the type will sit, reduce the palette to two or three colors. Most AI output is over-saturated in a way that reads as cheap.
- Set type over it. On its own layer, where you control size, spacing and color.
- Prefer texture over polish. Grain, brushwork and visible imperfection are currently doing well precisely as a reaction against the machine-made look.
And remove any text the generator put in the image — the fake quote, the invented imprint, the stray "A NOVEL". Clone it out or crop it away. Your title belongs on a live text layer, in a size and typeface you chose.
Declare it
Some retailers, Amazon KDP among them, ask you to state whether your cover or interior contains AI-generated content. Answer honestly. It is a checkbox in the upload form, not a scandal.
Make yours
Everything here is built into the tools, and they are free to use: name your book, bring your manuscript or your artwork, and work as long as you like. You only pay when you want the files without the preview watermark.
Next: Stock images: what the license actually lets you do · More: Artwork resolution for print · Composition and negative space · Mistakes that mark a book as self-published