What to Reject in an AI Cover Concept
Generating a cover concept is fast. Approving one should be slow. The image will often be confident and wrong: a handsome layout with a misspelled title, a face that resembles a real person, a genre costume that does not match the draft. A reject list turns “I don’t love it” into a reason you can act on. Without reasons, you regenerate at random and keep the loudest file.
ImprintaAI’s cover tool returns one concept image from the title, optional subtitle, genre, and style you submit. You review it. Nothing in that flow checks spelling against your draft, clears rights, or compares the image to your imprint rules. Those checks are the list below. KDP’s content guidelines still apply when you publish, including your responsibility for intellectual property and for an accurate presentation of the book. A rejected concept never has to meet that bar. An approved file does.
Reject unreadable or invented type
Read every word in the image, including words in the background, on signs, on book spines inside the picture, and in a signature scrawl. Reject the concept if any of these are true:
- The title is misspelled or missing a word that changes the name.
- Extra words appeared that you did not write.
- Letters merged, doubled, or turned into ornaments that are not letters.
- The author name is wrong.
- A subtitle promises a volume, a format, or a benefit you did not request.
Do not keep a wrong-title image as “the composition we like” unless you are prepared to rebuild the art without those pixels. People trace the letters by accident. If the type is baked into a texture, assume it is expensive to repair and reject early. The type hierarchy you wanted is easier to set in real fonts on clean art than to excavate from a painting of fonts.
Small type that is merely decorative and illegible is also a reject if a shopper could mistake it for a real subtitle. Fake Latin, fake reviews, and star rows are claims. Remove the image from consideration.
Reject anatomy and object errors that make the book look unfinished
Extra fingers, melted ears, teeth that blur into a texture, limbs that join the wrong body, and tools that could not exist are the usual tells. One oddity in a background can be cropped. An oddity on the focal point cannot. Puzzle-book covers that show a grid with letters that are not letters will bother the exact reader you want. Cookbook covers with a utensil that bends into the food have the same problem.
Reject distorted flags, maps, and instruments if the book’s subject depends on them being right. A field guide cover with the wrong bird is not a small error. It tells the reader the interior may be careless too. You can be more forgiving on a novel’s background smoke. You should not be forgiving on the one object the genre signal relies on.
Reject likeness and rights problems
Reject a face that looks like a specific living person, a celebrity, or you, unless you have a clear right to use that likeness and you meant to. Reject cartoon characters, superhero costumes, sports logos, and corporate packaging that a model dropped in because they are common in pictures. You do not acquire rights by generating a lookalike. The U.S. Copyright Office’s Copyright Basics is a general orientation, not advice about your file. When you are unsure, do not publish the image. Commission or photograph something you can explain.
Also reject compositions that trace a famous cover: the same crop, the same title placement, the same singular prop associated with one book. Genre habits are allowed. A photocopy of someone else’s design is not a habit. If you cannot describe the difference without naming another book’s cover, you do not have a difference yet.
Text from poems, lyrics, or other books inside the image is a reject unless you control that text. Models invent quotations that resemble real ones. Do not print them.
Reject genre drift and promise drift
Compare the image to the one-line promise on the cover brief. Reject it when the shelf is wrong: romance lighting on a workbook, gore on a cozy, a toddler palette on an adult large-print puzzle, a clinical cross on a cookbook. Also reject a promise the book does not make. A glowing brain on a journal does not make the journal a medical device, and it does invite that reading. Take it off. Money imagery that implies income does the same kind of harm.
Drift inside a series is its own reject. If book two’s candidate would not be recognized beside book one at thumbnail size, it fails even if it would be a fine standalone. The series rules outrank a one-off image you happen to like more.
Reject thumbnail failure
Run the thumbnail test as a hard gate. Title unreadable, contrast gone, or the wrong object remembered: reject. Do not create an exception because the full-size file is dramatic. Drama that shrinks into mud will not carry the imprint.
Keep a reject note so the next attempt is smarter
Write one line in the project when you reject a file. “Failed: title spelled Harbour, extra hand on the cup, reads as romance.” The next generation should change an input that addresses the note. If you cannot name an input the tool actually has — title, subtitle, genre, style — the change may belong in the brief, not in another click. Shorten the subtitle. Switch style from Bold and Colorful to Minimalist if the imprint is quiet. Change genre if you had picked a neighboring shelf by mistake.
Stop after a small number of attempts if the same failure repeats. That is a sign the idea is wrong, not a sign you need a rarer adjective. Take the brief to a designer, or simplify the cover to type on a field you control. A plain cover that says the true title is a better package than a striking cover you should have rejected.
Approval, when it comes, gets a line too. “Kept: title readable at 160 pixels, one garden tool, no extra words, matches book one title block.” That sentence is the standard the next volume has to meet. Rejection is how the standard stays real.