QWE AI Academy Your AI-generated book cover looks perfect on screen. Then Amazon rejects it.
The problem isn’t the AI – it’s the gap between generation and upload. Most tutorials stop at “download and add text in Canva.” But between that glossy PNG and a printed book sits a technical minefield. 80% of AI covers die here.
The Resolution Trap: Why Your Screen Lies
Your monitor shows images at 72 DPI. “Dots per inch” – the density of visual information. At 72 DPI, a Midjourney image looks sharp.
Print needs 300 DPI.
Midjourney’s default: 896 by 1344 pixels. Portrait-oriented, looks right for a book. But Amazon KDP requires minimum 1600 by 2400 for ebooks, print books need 1800 by 2700 at 300 DPI (as of 2025). Your 896-pixel image? Half the required width.
Upscaling fills the gap with computer-generated pixels – resampling. Sometimes it works. Often? Blurry edges. Lost detail. Color shifts. Amazon’s automated QA scans for this. Low resolution detected → rejected before a human sees it.
You won’t know until upload.
What Actually Breaks (and When)
Three failure modes:
DPI doesn’t match physical dimensions. Resolution ties to print size. A 3-inch photo at 300 DPI becomes 1.5 inches if you double DPI without changing pixel count. Or 6 inches at 150 DPI – too low. Most AI tools output at 72 DPI. Built for screens, not paper. Converting 72 DPI to 300 DPI doesn’t add detail. The software guesses what missing pixels should look like.
Text generation is broken. AI models fail at typography – garbled letters, misspelled words, fonts that shift mid-word. Amazon’s guidelines prohibit “blurry, pixelated, or illegible text”. Any AI text attempt – even decorative flourishes – risks rejection. Community data (as of 2025): ~30% of AI generations include text distortion, even from Midjourney v6.
Copyright ambiguity isn’t theoretical. January 2025: U.S. Copyright Office ruled purely AI-generated works can’t be copyrighted. Prompt-only generation? No human editing? Anyone can copy your cover legally. You can’t stop them. Canva’s terms warn: “you may not have exclusive rights to your book cover images.”
Think about that. You spend an hour perfecting a cover. Someone screenshots it from your product page. Uses it on their book. Legal.
The Workarounds That Actually Work
Resolution fix needs two steps. First: generate larger. Request maximum size if your tool allows it. Midjourney’s upscale function helps – but it’s still resampling. You’re smoothing what’s there, not creating detail.
Second: dedicated upscaling service. Free options like Upscale Media. AI-powered tools like Topaz Gigapixel. They analyze the image, reconstruct missing information. Not perfect – better than Photoshop’s bicubic resampling. Start with the highest-quality generation your tool allows. Then upscale. Can’t add detail that was never there.
Typography? Simple rule: AI does imagery. You do text manually. Canva. Adobe Express. PowerPoint even. These give you typographic control AI can’t match. Serif for literary fiction. Bold sans-serif for thrillers. Script for romance. Text is where readers decide to buy. Don’t let a robot do it.
Copyright protection needs human input. Copyright Office test: “sufficient creative control over expressive elements.” Typing a prompt doesn’t count. Generate base image, edit in Photoshop – adjust lighting, composite elements, paint over sections – that human work makes it copyrightable. More you modify, stronger your claim. Some designers use AI for individual elements (character, background, prop), manually compose into final cover. Speed of AI + legal protection of human authorship.
The Paid Tool Trap
Free tiers make your generations public. That fantasy castle you perfected for an hour? Gallery somewhere. Tagged with your prompt. Visible to anyone – including competitors. Serious about AI covers? Budget for paid plans. Midjourney basic: $10/month (as of 2023, pricing may have changed). DALL-E charges per generation, allows commercial use. Privacy is worth it.
Testing Before You Commit
I generated a cover. Midjourney. Prompt: “dark fantasy book cover, ancient library, glowing runes, atmospheric lighting –ar 2:3”. Initial: 1024×1536 pixels. Better than default, still under KDP minimum. Upscaled with Topaz Gigapixel to 2400×3600. Added title and author in Canva – Cinzel font, matches fantasy aesthetic.
Uploaded to KDP as draft. Passed automated check. Ordered print proof.
Cover looked sharp in thumbnail. Physical book? Runes showed slight blur. Upscaling artifact. Readable but not professional.
Second attempt: generated at max resolution (Midjourney beta high-res mode). Upscaled conservatively. Simplified background detail – less strain on upscaling. Proof copy was cleaner. Runes stayed sharp. Worked because I knew where it breaks.
When AI Isn’t the Right Tool
Romance covers need photographic models – specific poses, expressions. AI struggles with realistic faces. Especially hands. Distorted fingers, uncanny eyes, asymmetrical features. Stock photo compositing: faster and cleaner.
Literary fiction uses minimalist, typographic covers. AI adds no value.
Children’s books need consistent characters across pages. AI can’t generate the same character twice reliably. Ideogram’s improving – not there yet (as of 2025).
AI’s sweet spot: genre fiction that focuses on setting over character. Fantasy landscapes. Sci-fi tech. Thriller atmospherics. Horror moods. If the cover sells a vibe rather than a person, AI works.
What Changes Next
Models improve fast. DALL-E 3 does text better than DALL-E 2. Midjourney v6: higher default resolution than v5. But fundamentals – print DPI requirements, copyright ambiguity, Amazon’s quality bar – won’t change. Understanding technical requirements matters more than chasing the newest tool.
One open question: hybrid workflows. Some designers now use AI for concepts, hire illustrators to refine. AI does ideation. Human finishes execution. Sidesteps copyright (final work is human-authored), keeps costs lower than full custom. Not what tutorials recommend – might be where the industry settles.
FAQ: AI Book Cover Problems
Can I just upscale a Midjourney image in Photoshop and stop there?
Photoshop’s default resampling (bicubic) smooths existing pixels – doesn’t add detail. For print: need larger source or dedicated AI upscaler (Topaz Gigapixel, free Upscale Media). Test first. Order print proof before full run.
If I edit an AI-generated cover in Photoshop, do I own the copyright?
Comes down to how much you edit. U.S. Copyright Office (January 2025): need “sufficient creative control over expressive elements.” Light color correction? Probably not enough. Compositing multiple AI elements, repainting sections, designing layout from scratch? Probably counts. The line’s fuzzy. Untested in court. If copyright matters for your project, talk to a lawyer. Or skip AI, commission original art. The risk: someone copies your cover, you sue, judge rules your edits weren’t substantial enough. You lose – and paid legal fees for nothing.
Why does my AI cover look great on my laptop but blurry in the KDP preview?
Screen: 72 DPI. KDP preview: shows print at 300 DPI. If source file lacks pixel density, preview reveals the problem. Check image properties. Under 1600×2400 pixels? Too small. Don’t trust your screen. Check real pixel count before uploading.