Kdp Three Artifacts Html Description Cover Type Brief

Kindle Direct Publishing does not need a vibe. It needs three files that survive the form: an HTML product description, a 2:3 cover image, and a type brief so the title on that cover is still the title you wrote.
Most KDP prompt stacks try to write the whole book, then "also make a cover." That is why the listing reads like a movie trailer, the cover has a different subtitle, and the type is unreadable at thumbnail size. Split the job. Matching KDP artifact prompts are landing in the PromptDig browse library (Browse more prompts).
Artifact 1: the HTML description, not a blurb
The KDP description field accepts a small HTML subset: bold, italic, lists, line breaks. It does not accept your Google Doc. If you paste markdown or a wall of marketing adjectives, you will spend the afternoon in the previewer.
Write the description as HTML on purpose. Prompt for tags, not for "an exciting summary."
Write a KDP product description in HTML only.
Allowed tags: p, br, b, i, ul, li, h4.
Structure:
1) One-paragraph hook. No rhetorical questions.
2) Bulleted list of what the reader gets (3 to 5 items).
3) Who it is for / who it is not for.
4) One-line proof the reader can verify (edition, what is inside).
Do not invent reviews, rankings, "bestseller" claims, or reader counts.
Do not mention Kindle Unlimited unless I say it is enrolled.
Plain title case for the book name: [title].
The "do not invent reviews" line is the whole game. Models love a fake testimonial. KDP and readers do not.
Keep the description scannable on a phone. Short paragraphs. A list.
Save the HTML in a file. Paste from the file. If you keep rewriting it in the KDP box, you will lose the tags.
Artifact 2: a 2:3 cover that still works small
Ebook covers on Amazon are judged at thumbnail size first. A 2:3 canvas (for example 1600 x 2400) is the ratio you actually see in the grid. Square "book mockups" from a generic image prompt will look like posters.
Prompt the image model for a cover, not a scene:
- 2:3 portrait, full-bleed
- Large clear area for title (top or center, not both)
- No fake author names, no fake awards, no fake barcodes
- No tiny texture that dies at 150 pixels wide
- Subject that reads as a silhouette: one object, one landscape band, one figure
Then add the lock that image models ignore unless you shout it: "Leave the title area empty. Do not render letters. Typography will be added in a layout tool."
If you let the model invent the title, you will get mushy letterforms and a subtitle you never wrote. Generate the background. Set type yourself, or in a tool that outputs real text.
For a paperback wrap you need a different file with a calculated spine width. Do not stretch the 2:3 ebook front into a wrap. Keep the ebook front as the master.
Artifact 3: the type brief, so the cover and the listing agree
The type brief is the missing artifact. It is a one-pager the cover designer (human or model-plus-layout) cannot violate.
Minimum fields:
- Title, exactly as it will appear on KDP (punctuation, small caps, ampersand or "and")
- Subtitle, or an explicit "no subtitle"
- Author name as it should be letterspaced
- Hierarchy: title size feel (big / medium), subtitle smaller, author smallest or reverse for a name-driven cover
- Type style in plain language: high-contrast sans, literary serif, condensed display. Not "elegant" or "modern"
- Color: title color that still reads on the background, plus a fallback if the background is busy
- Do not: script on a texture, more than two type families, award seals, fake series numbers
A short prompt that produces the brief instead of a new cover:
From this title, subtitle, and cover description, write a type brief.
Output: exact strings to set, hierarchy, suggested type pairing in words,
safe color for type, and a thumbnail test: "at 100px wide, only the title must read."
If the subtitle fights the title, recommend dropping the subtitle on the cover
and keeping it on the KDP form only.
Do not generate an image. Do not invent a different title.
Run that brief before you generate the 2:3 image, not after. The empty type area on the cover should match the hierarchy you just chose. A big serif title needs a quiet sky or a flat color block. A condensed sans can sit on a busier photo.
One pass through all three, then stop tinkering
Order of operations:
- Freeze the title string. If it is still a working title, do not make a cover.
- Write the type brief.
- Generate the 2:3 cover with an empty type area. Set real type.
- Write the HTML description using the same title string and the same promise as the cover.
- Check the thumbnail, the KDP preview, and the description on a phone.
If the cover promises a thriller and the HTML description sells a workbook, you made two books. Pick one. Change the artifact that is lying.
Keep the three files together: description.html, cover-2x3.png, type-brief.md. That bundle is what you reuse for a sequel or a second edition. The manuscript can change. The bundle should still match.
When a stack finally produces a listing that matches its cover, share the three prompts as a set (Share a prompt). The adjectives in the blurb are cheap. The HTML rules, the empty type area, and the exact title string are the asset.
KDP is a form with opinions. Give it three artifacts it can accept, not a mood board.