AI Cover Art for Authors: What KDP and ACX Require and What Image Tools' Terms Say

If you use AI to make a book cover, KDP treats the image as AI-generated content you must tell it about, and that stays true however much you edit it afterward. This is general information, not legal advice. Check the image tool's own terms too, because we could read only some of them.

KDP's rule on AI images

KDP's AI content page says you must inform KDP of AI-generated content (text, images or translations) when you publish a new book or edit and republish an existing one. It says content created by an AI-based tool is AI-generated "even if you applied substantial edits afterwards", and that covers and interior images count as AI-generated images. AI-assisted content, where you created the work and used AI to edit or refine it, need not be disclosed. KDP also says you are responsible for verifying that AI-generated and AI-assisted content complies with its guidelines and with intellectual property rights. The page does not set out how to judge where a heavily retouched image falls, so see the KDP disclosure guide.

Here is how that wording applies to common cover situations. This is our reading of the page, not a statement from Amazon.

Cover specs side by side

One image rarely fits all three formats, so check the specs before you generate anything. KDP and ACX figures are from each page as read on 2026-11-08.

Sources: KDP ebook cover criteria, KDP paperback cover page, and ACX's cover art requirements. ACX's AI rule concerns narration, not covers, and this page does not say whether an AI-generated cover is accepted there. Cover specs for other audio retailers were not read; see audiobook audio specs. For trim sizes and interior bleed, see print book formatting requirements.

KDP also adds a barcode to a paperback if you omit one, and it warns against borders because of production variance.

A worked paperback size (our arithmetic, an estimate)

KDP's formula is bleed + back + spine + front + bleed for width, and bleed + trim height + bleed for height, with bleed of 0.125 inches. Take an invented 6 x 9 inch paperback of 300 pages on white paper, where KDP's spine figure is 0.002252 inches a page.

• Spine: 300 x 0.002252 = 0.6756 inches.
• Width: 0.125 + 6 + 0.6756 + 6 + 0.125 = 12.9256 inches.
• Height: 0.125 + 9 + 0.125 = 9.25 inches.
• At 300 DPI: about 3,878 x 2,775 pixels for the full wrap.

Page count changes the spine, so finish the interior first. KDP's page also links a cover calculator and template generator; use that for the real file.

One master image for all three (our arithmetic, an estimate)

If one painting has to serve all formats, the front panel of that paperback needs 6.125 x 9.25 inches, or about 1,838 x 2,775 pixels at 300 DPI (trim width plus outer bleed, by full height). ACX needs 2,400 pixels of width, and KDP's ideal ebook needs 2,560 of height. So a master of at least 2,400 x 2,775 pixels covers all three. Keep the subject in the central square, because ACX will crop to it. Check the pixel size your tool actually exports, and do not assume it; enlarging a smaller file does not add detail.

A cover brief you can fill in

A brief forces the decisions that a prompt tends to skip, and it works whether you use a generator or hire a designer. Copy this and fill in each line.

Title and subtitle:
Author name, exactly as credited:
Genre, and three comparable covers you have looked at:
One-line promise of the book:
Main image, in one sentence:
Mood, in three words:
What must not appear:
Formats needed: ebook / paperback / audiobook
Trim size and page count:
Where the title will sit, and how much clear space it needs:
Tool or designer, and date you read its terms:
Text added by: (a person, in an image editor, not the generator)
KDP category: AI-generated / AI-assisted / neither

An invented example for a book that does not exist:

Generate only the art. Set the title and author yourself, since each retailer's text rules apply to the final file, and letters drawn by a generator need checking character by character.

Image tool terms: read and unread

• Ideogram, read. Its Terms of Service, last revised August 14, 2024, say "We do not claim any ownership rights in your User Input or User Output" and allow commercial use. They also disclaim liability for commercial use, say output may not be unique, say others may remix your images unless private, and tell you to consider what permissions you may need from third parties. The terms were silent in our read on whether a paid plan is needed for commercial use.
• Adobe Firefly, partly read. Adobe's Generative AI User Guidelines, updated May 15, 2026, are silent on ownership and commercial use, and say you must not remove, alter or disable Content Credentials. Adobe's commercial-use terms were not read.
• Midjourney and Canva, not read. Their terms pages returned access errors on 2026-11-08. We state nothing about ownership or commercial rights for either. Read their terms before you publish.
• OpenAI's image tools, not read. OpenAI's terms returned an access error on 2026-11-08.
• Other generators. We did not attempt Stable Diffusion, Leonardo or other tools' licences for this page.

Summaries of these terms on other sites are not a substitute, and terms change, so note the date you read them.

When you read a tool's terms, look for five answers: who owns the output, whether your plan permits commercial use, whether others can see or reuse your images, what the vendor disclaims, and what happens to those rights if you cancel.

Copyright

We read the US Copyright Office's Part 2 report (January 2025) from the PDF text. It is the Office's analysis of US law, not a court ruling, and it does not discuss book covers as such. Four points bear on them.

• Prompts. The Office concludes that with current generally available technology, prompts alone do not provide sufficient control, and that repeatedly revising a prompt does not change this. It describes re-prompting as re-rolling the dice.
• Your own input. If you put your own copyrightable artwork in and it is perceptible in the output, you are the author of at least that portion. The Office registered one such image, limited to the human drawing. The report contrasts a refused case where a photograph was the input and the output did not clearly show it.
• Modification and arrangement. Creative changes and creative selection or arrangement can be protected, but the protection does not extend to the AI-generated material standing alone. Whether a change is enough is decided case by case.
• A limit worth knowing. The report's footnote 138 says arranging only two or three elements is not generally enough for protection as a compilation. So do not assume that placing a title on a generated image creates a protected arrangement. The report does not rule on that exact case.

See our AI and copyright overview and the Copyright Office AI page.

Practical checklist

1. Read the image tool's terms for commercial use before you generate.
2. Fill in the brief, including what must not appear.
3. Save the date, plan and prompt for your records.
4. Inspect the image at full size. Check hands, faces, text, repeated textures and anything that resembles a real person, a logo or a known character. This is general advice, and KDP puts responsibility for intellectual property on you.
5. Set the title and author yourself, and proofread them.
6. Check the final file against each retailer's spec above.
7. Disclose in KDP if the image is AI-generated.
8. Keep Content Credentials intact where a tool adds them.

Who should skip it

If you need copyright in the art itself, note the Office's analysis that purely AI-generated material is not protected, and ask any designer to state in writing which elements an AI tool created. Do the same if your readers or your contract object to AI art, or if you are unwilling to tell KDP. We have no data on how AI covers sell, and we have not compared them with designer work.

More tools and guides are in the AI drafting tools hub.

Last verified 2026-11-08. Rules and prices change: check the linked primary page before you rely on this.