Self-publishing on Amazon KDP in 2026 produces more books than the traditional publishing industry. Most of them look amateurish. Their covers signal "indie" in the worst way: metaphor-literalism, over-designed elements, motivational stock imagery, font-chasing. And most importantly — they don't convert at Amazon's thumbnail size, which is where nearly all discovery happens.
This guide is the honest playbook for using AI (specifically ChatGPT Images 2.0 / gpt-image-2, plus complementary tools) to design covers that actually sell books.
It covers:
- Why most AI book cover generators produce covers that fail at Amazon thumbnail size
- Genre-specific conventions that readers use to filter (and therefore your cover must match)
- The typography hierarchy that separates $500 custom covers from free AI attempts
- How to use ChatGPT Images 2.0 vs Midjourney vs Imagen 4 for different cover needs
- Post-AI workflow in Figma for production-ready covers
- A/B testing via Amazon Ads (the only honest way to know if your cover converts)
- When to hire a real designer (and when AI-generated covers are genuinely good enough)
---
The Thumbnail Reality That Kills Most Indie Covers
Amazon shows book covers at 100×150 pixels in search results and category browse pages. That's tiny. If your title isn't legible at that size, nothing else matters — readers scroll past.
Test yourself right now: open Amazon's book search, look at the thumbnail grid. Which covers stop your scroll? Almost all of them have:
- A clear title legible at thumbnail size
- An iconic visual element (or typography-dominant design)
- Genre conventions matched (thriller looks thriller, romance looks romance)
- Simple, bold color signaling the mood
- Author name present but not competing with title
What kills most indie covers:
- Title too small — readable on the full-size PDP page, invisible at thumbnail
- Multiple visual elements competing for attention
- Gradient backgrounds (AI tools love gradients; book designers don't)
- Motivational stock imagery (mountains, arrows, silhouettes)
- 3+ fonts used
- Metaphor-literalism — title says "The Lean Ship," cover has a literal ship
AI book cover generators produce all of these problems by default. Good outputs require specific prompting discipline.
---
Genre Conventions: The Filter Readers Use
Readers filter covers unconsciously in under two seconds. If your cover doesn't signal the right genre, readers skip it — they're looking for thriller, they see romance, they move on.
Commercial Fiction Conventions
Thriller / Mystery / Crime:
- Dark atmospheric color palette (blacks, deep blues, crimson accents)
- Single iconic visual (running figure, silhouette, mysterious object)
- Bold serif or sans-serif title, high contrast
- Minimal additional elements
- Examples: Gone Girl, The Girl on the Train, The Silent Patient
Romance:
- Warm color palettes (pinks, warm oranges, soft yellows)
- Couple or romantic scene illustration
- Script or cursive title typography
- Visible author name (readers follow romance authors)
- Sub-genres matter further (contemporary vs historical vs paranormal)
- Examples: Emily Henry books, Colleen Hoover, Helen Hoang
Sci-Fi / Fantasy:
- Cool metallic palettes (silvers, deep blues, purples) or epic warm tones
- Scene-driven with epic scale (landscapes, creatures, characters)
- Fantasy-specific serif fonts (Cinzel, Trajan, Optimus)
- Dramatic composition
- Examples: Brandon Sanderson covers, N.K. Jemisin, Sarah J. Maas
Literary Fiction:
- Muted palettes, sophisticated color choices
- Understated imagery
- Elegant serif typography
- Minimal additional elements
- Examples: Sally Rooney covers, Ocean Vuong, Yaa Gyasi
Non-Fiction Conventions
Business / Startup:
- Bold confident color palette (navy, red, orange, deep green)
- Typography-dominant (title does the work)
- Minimal illustration if any
- Sans-serif weight-heavy fonts
- Examples: The Lean Startup, Atomic Habits, Deep Work
Self-Help:
- Bright optimistic colors or confident earth tones
- Iconic symbol or minimal illustration
- Bold sans-serif title
- Benefit-promise subtitle prominent
- Examples: Mindset (Dweck), Quiet (Cain), Daring Greatly (Brown)
Memoir:
- Muted warm colors
- Intimate portrait or specific personal imagery
- Serif typography (literary feel)
- Understated, personal
- Examples: Educated (Westover), When Breath Becomes Air, Maid
Cookbook:
- Bright food-forward imagery
- Clean modern typography
- Author name prominent (readers follow chefs/cooking personalities)
- Category signaling (Italian, vegan, weeknight, etc.)
- Examples: Salt Fat Acid Heat, Half Baked Harvest, Ottolenghi
Matching your genre's conventions isn't constraint — it's the first filter. Then differentiate within the genre.
---
Typography Hierarchy: What Actually Matters
Cover design professionals obsess over typography hierarchy because it determines whether readers can process the cover at a glance. AI tools get typography fundamentally wrong by default.
The Rule: Title > Subtitle > Author (usually)
Fiction: Title dominates. Author name smaller. Subtitle often nonexistent.
Non-Fiction: Title + Subtitle often equally important because subtitle contains the benefit promise. Author name smaller.
Memoir: Title typically prominent, but author name visible at similar size because readers often connect with specific authors.
Title Treatment Specifics
For commercial fiction thriller: bold sans-serif or display serif, 70-80% of cover's top 60% space.
For literary fiction: elegant serif, often centered, moderately sized.
For non-fiction business: bold weight sans-serif, dominant at top 60% of cover, often all caps.
For romance: script or decorative typography, often with subtle flourishes.
For sci-fi / fantasy: genre-specific fonts (Cinzel, Trajan, Optimus, Immortal). Display serifs with dramatic proportions.
The Subtitle Load (Non-Fiction Specifically)
Your subtitle is SEO + reader-qualification. It tells readers in two seconds whether this book is for them.
Weak: "How to Succeed in Business"
Strong: "The Practical Guide to First-Time Founders Building Products Customers Actually Want"
The strong version:
- Says WHO it's for (first-time founders)
- Signals WHAT (practical guide, not theory)
- Promises BENEFIT (products customers actually want)
AI tools default to weak generic subtitles unless you specify exactly what you want.
---
ChatGPT Images 2.0 for Book Covers (Specific Tactics)
ChatGPT Images 2.0 (gpt-image-2) launched April 2026 with significantly improved text rendering — the first image model where "book cover with actual readable title" is reliably achievable.
What gpt-image-2 does well for book covers
- Readable title typography at moderate sizes
- Genre-aesthetic capture (thriller moody, romance warm, etc.)
- Composition and layout following instructions
- Color palette adherence
- Multi-panel coherence for series covers
What it still fails at
- Exact brand font reproduction (custom fonts you own → not possible via AI)
- Small type at fine detail (subtitle at 6-8pt may have rendering issues)
- Exact logo reproduction (don't try)
- Precise kerning for typographically-demanding work
- Specific real author photo (always redo in Figma for production)
The Book Cover Prompt Template
```
Book cover design for a [GENRE] [FICTION/NON-FICTION].
TITLE: '[EXACT TITLE]' (spell it out letter by letter if necessary)
SUBTITLE: '[EXACT SUBTITLE]' (or "no subtitle")
AUTHOR: '[EXACT AUTHOR NAME]'
Genre signaling: [specific to your genre — thriller/romance/sci-fi/business/etc.], comparable to [2-3 specific bestsellers in your sub-genre]. Explicitly NOT a [wrong genre] cover.
Visual approach: [typography-dominant / scene-driven / iconic symbol]. [Specific visual element if any, abstract otherwise].
Color palette: [specific colors with hex codes if known]. [Mood: moody / warm / confident / literary].
Composition:
- Top 60%: title treatment with [specific typography style]
- Middle 20%: subtitle (if applicable)
- Bottom 20%: author name
Optional small visual element: [specific iconic element if any], not dominant.
Feel: [aesthetic reference to specific book, designer, or style].
Format: portrait 6×9 (standard trade paperback). Clean margins. Amazon KDP suitable.
No endorsement quotes, no bestseller badges, no tagline beyond subtitle.
```
Genre-Specific Variations
Thriller Prompt Addition:
```
Dark atmospheric composition. High contrast. Single iconic visual element. Mood: moody, dangerous, tense. Think David Fincher film poster meets Gillian Flynn book aesthetic.
```
Romance Prompt Addition:
```
Warm color palette (blush, cream, soft orange). Script or decorative title typography. Romantic scene or couple illustration (silhouettes acceptable, detailed illustration preferred if style allows). Visible author name at similar weight to title.
```
Business Book Prompt Addition:
```
Typography-dominant. No illustration — title does all the work. Bold sans-serif in [color] on [background]. Subtitle in lighter weight, same font family. Confident, practical, not corporate. Think The Lean Startup meets Atomic Habits aesthetic.
```
Memoir Prompt Addition:
```
Muted palette, intimate feel. Serif typography (literary credibility). Subtle personal imagery OR typography-only. Think Educated by Westover meets When Breath Becomes Air. Understated, not commercial.
```
---
When to Use Other Tools Instead
Imagen 4 (Google)
Better than gpt-image-2 at:
- Dense typographic work (multiple text elements requiring precise kerning)
- Small type rendering
- Regulatory-compliant label typography
Use for: Covers with complex multi-line text, academic/professional covers with formal typography requirements.
Midjourney v7
Better than gpt-image-2 at:
- Specific aesthetic replication (1950s Penguin covers, specific film posters, specific artist styles)
- Photographic cover imagery with precise film-stock looks
- Fantasy/sci-fi scene composition with epic scale
Use for: Literary fiction with aesthetic-precise references, fantasy with specific atmospheric quality, covers where the visual mood is paramount.
Hand-designed in Figma (recommended for production)
Always do this regardless of which AI tool produced the base. The workflow:
- Generate 5-10 AI variations across different prompts
- Pick the best 2-3 compositions
- Recreate in Figma with your chosen brand typography (precise fonts, exact kerning, print-ready CMYK colors, proper bleed)
- Composite any real elements (author photo for back cover, specific objects)
The AI produces the concept. Figma produces the production-ready file.
---
The Amazon KDP Production Workflow
1. Design
Cover dimensions: 1600×2560 pixels minimum for Amazon KDP ebook. Print version requires additional bleed (0.125" on all sides for paperback, 6.125×9.125 effective).
2. Color Space
RGB for ebook (displays on screens). CMYK for print (KDP converts but working in CMYK from start produces more predictable print output).
3. Spine
Paperback books need spine design based on page count. KDP's spine calculator gives exact width. Spine width affects whether text and branding fit properly.
4. Back Cover
Short book description (150-250 words), author bio (2-3 sentences), optional author photo, ISBN barcode space (KDP provides automatically).
5. File Format
PDF with embedded fonts, 300 DPI for print, RGB 72 DPI acceptable for ebook.
6. Preview Testing
Upload to KDP preview BEFORE publishing. See how it renders in Amazon's viewer. Thumbnail test at 100×150px — critical.
---
A/B Testing Your Cover
The only honest way to know if your cover converts is testing it in the market.
Amazon Ads A/B Test Setup
- Create two distinct cover variations (not minor variants — genuinely different design approaches)
- Upload both to KDP
- Run Amazon Ads for each cover: $50-100 per variant over 7 days, same target audience
- Measure: click-through rate from ads, conversion rate to purchase
Lower cost-per-click + higher conversion wins. Statistical significance usually visible at 500+ impressions per variant.
What to Test
- Typography-dominant vs scene-driven design
- Different color palettes within genre range
- Different title treatments (bold uppercase vs elegant sentence case)
- With vs without author photo
- Different aesthetic references (noir vs classic thriller aesthetic)
What NOT to Test
- Minor variations (different shade of same color, one font vs another very similar font)
- Price points (separate test category)
- Book descriptions (test separately)
---
When to Hire a Real Designer
AI-generated book covers are now genuinely good enough for many use cases. But some situations still warrant professional design:
Hire a Designer If
- Genre-specific visual aesthetic that AI can't nail (high-concept literary fiction, complex illustrated children's books)
- Series cover system that needs perfect consistency across 3+ books
- You're investing significant marketing budget — design investment scales with marketing investment
- Your first impression matters massively (debut author, major career book)
- You're partnering with traditional-style publishing (retail bookstores expect certain design quality)
Budget Ranges (2026)
- Reedsy.com marketplace designers: $300-1,500 per cover (good quality range)
- Damonza: $200-600 (specialized genre designers)
- Individual freelancers on Upwork / 99designs: $150-500 (wide quality range)
- Top-tier independent designers: $1,500-5,000 (when quality is paramount)
AI Is Good Enough When
- You're testing a new genre or sub-genre (low commitment)
- Budget is very limited
- You have Figma skills to refine AI output to production
- Series books 2+ in an established template you've already committed to
- Genre conventions are relatively simple (standard thriller, romance, self-help)
---
The Promptolis Book Cover Designer Pack
We maintain a Book Cover Designer Pack with 20 prompts calibrated to genre-specific conventions:
- Fiction thriller / mystery templates
- Romance (contemporary + historical)
- Sci-fi / fantasy scene composition
- Literary fiction typography-first
- Business / startup / productivity
- Memoir / narrative nonfiction
- Children's book (age-specific)
- Series cover template systems
Each prompt includes the ChatGPT Images 2.0 template, variant genre adaptations, post-generation Figma workflow, and Amazon KDP production specifications. Research-backed by Joel Friedlander (The Book Designer), Reedsy cover design research, and Ellen Lupton's typography work.
Free, MIT-licensed. Built for self-publishers who ship.
---
Common Mistakes Self-Publishers Make
1. Metaphor-Literalism
Your title says "The Lean Ship" (business metaphor). You want a literal ship on the cover. Don't. Business book covers are typography-dominant. Save the nautical imagery for fiction.
2. Too Many Elements
Typography + illustration + background texture + color gradient + effect overlay. Pick ONE visual approach. Typography-driven OR single iconic illustration OR atmospheric scene. Not all three.
3. Motivational Stock Imagery
Mountain-climbing silhouettes, arrows pointing up, professional in suit, hands raised to sky. These signal "2010 self-help book" immediately. Avoid unless you're writing specifically in that category.
4. Fake Professionalism Font-Chasing
Using five different fonts to look "designed" produces the opposite. Professional covers use one font family in 2-3 weights maximum.
5. Ignoring Thumbnail Testing
Your cover looks great at full size. It's illegible at 100×150. Amazon is a thumbnail-first environment. Test in preview BEFORE publishing.
6. Skipping the Genre-Comp Analysis
Not knowing what the bestsellers in your genre look like. Before designing, spend 30 minutes studying Amazon's top 100 in your specific sub-category. Note patterns. Match the conventions. Then differentiate within them.
7. Series Consistency Abandoned
Book 2 looks nothing like Book 1. Series readers want visual consistency across all books. Design a template for Book 1 that Books 2-10 follow rigidly.
---
What to Expect
If you follow this guide:
- Your cover will match genre conventions (filter passed)
- Typography will be legible at thumbnail (first hurdle passed)
- Visual approach will signal the right mood (second hurdle passed)
- Title + subtitle will work together (qualification done)
Then Amazon's algorithms can actually do their work — showing your book to readers who might buy it.
What this guide can't do:
- Make a bad book succeed (cover gets the click; book quality drives reviews and sustained sales)
- Guarantee marketing success (you still need Amazon Ads, reviews, email list)
- Replace craft (a great cover for a poorly-written book produces disappointed readers)
Good cover + good book + good marketing is the trifecta. Missing any one collapses the other two.
---
Get Started
- Research your genre comps (30 minutes)
- Draft your title + subtitle + author name
- Use the Book Cover Designer Pack prompts in ChatGPT Images 2.0
- Generate 5-10 variants
- Refine in Figma for production
- Thumbnail test at 100×150
- A/B test via Amazon Ads if investing marketing budget
Six weeks from concept to published cover is very achievable with AI tools in 2026. Six years ago the same path required $500+ designer fees and multiple iterations. The accessibility is new. The craft fundamentals aren't.
Write good books. Design real covers. Ship consistently.
— Atilla