Carousel modes, explained: per-slide vs single-repeated, slide caps, and per-slide overlays

If you opened the campaign builder a few months ago, every carousel was the same shape: AI generates one image per slide, you got 5-10 slides, you posted it. Done.

That's not how carousels work in the tool anymore. Two image strategies, a recipe picker, per-platform caps, and a real panel text editor — all surfaced in the campaign wizard. Picking the right combination for a specific batch is now a real decision. Here's how to think about it.

The single most important change first: the words on your panels are typeset by the app in real fonts, never painted by the image model. Image models can't spell; the app can. Your carousel image prompts describe wordless backdrops, and the panel editor lays your actual text over them — so a misspelled quote card is a thing of the past.

Two modes: per-slide and single-repeated

Every carousel post has a mode that determines how slides relate to each other.

Per-slide ("a fresh image every panel" in the wizard) is the classic carousel: each slide has its own image and its own overlay text. Five slides = five distinct images. Use this when each slide is a different beat — character reveals are the flagship case, and the Character Reveals recipe pairs with it naturally.

Single-repeated ("one background, words change" in the wizard — the default) uses one image across all slides: the same atmospheric backdrop is the canvas for different text per swipe. Five slides = same image, five different text layers, one image spend. Use this when the words are doing the storytelling. Good for: book quotes, trope cards, POV/hook escalations, an "if you liked X, try Y" list laid over a single aesthetic.

Pick the strategy as tickboxes on the wizard's Basics step, right under the slide count.

Recipes: the formats readers actually swipe

Also on Basics: a recipe picker with four proven carousel formats — Quote & dialogue (lines from the book escalating swipe by swipe), POV / hook escalation (panel 1 stops the scroll, every swipe raises the stakes), Trope card ("This book has:" — one searched-for trope per panel), and Character reveals (one character per panel, each with its own art). Pick several and the batch spreads across them; pick none and the plan mixes all four.

Two automatics worth knowing: every carousel ends on a call-to-action panel, and when a book is attached to the campaign, that closing panel is your real cover composed on a blurred atmosphere — no AI spend, no mangled cover art. If you've saved quotes to the book's reference library, the Quote & dialogue recipe uses your actual lines verbatim instead of inventing any.

Slide count: 2 to 10

The slides-per-carousel field on the Basics step takes anywhere from 2 to 10. Most authors land on 4-5.

Practical guidance:

• 3-5 slides — for skim-friendly content. Hook, one or two body slides, CTA. Highest completion rate.
• 5-7 slides — the sweet spot for most narrative carousels. Hook, 3-5 body beats, payoff, CTA.
• 8-10 slides — instructional or list content where each item needs its own slide. Drop-off increases past slide 8 — make slides 1-3 strong enough to hook even if not everyone makes it to the end.
• 10 — the cap; only for genuinely list-heavy content. Most carousels don't need it.

If AI writes fewer slide texts than you asked for, the tool pads up to your requested count. If it writes more, it trims. Either way you get exactly the slide count you picked.

Per-platform slide caps

Different platforms have different maximum carousel sizes. The tool enforces these at schedule time so you don't ship a 10-slide carousel to a platform that only allows 4.

The wizard warns you on Basics, before anything generates: platforms that cap the deck show what trims, and platforms that can't publish carousels at all are called out with the reason.

The panel text editor

Every carousel post in the Media studio carries a panel editor. Each panel has its own text; the deck-wide controls set the typography once for every panel — font (six bundled faces with live preview), color, size, horizontal alignment, vertical placement (top / middle / bottom), a background dim for legibility on busy images, and an optional solid text pill in any color. Everything previews live on the panel tiles as you type, autosaves as you go, and "Save text & badges to slides" bakes it into the images in real type. Each save starts from the clean original image, so you can re-edit endlessly without text stacking on text.

There's also a badge row — the corner stamp readers know from BookTok. Pick a text-pill preset (Kindle Unlimited, Free on Kindle Unlimited, Out now, Preorder now, 99¢ deal), type your own, or upload your own logo image (the official KU badge from your KDP marketing assets, say), choose the corner, and it stamps every panel.

Upload-Your-Own

Batches created with the "upload my own media" source skip AI backdrops entirely — you supply the images and the panel editor still handles all the text and badges. On AI batches, any panel can be regenerated alone if it doesn't land.

Use cases:

• The first slide is your actual book cover — generate the rest, upload yours for slide 1
• One slide needs a real photo (you, your workspace, an in-store shot) the AI can't produce
• A specific aesthetic you've already designed in Canva — paste the export in
• Mid-carousel content moment that needs to be exactly right (an author photo, a fan's repost screenshot, a chart)

And remember the closer is handled for you: with a book attached, the final panel is the real cover — you never have to upload it.

Cross-platform behavior in single-repeated mode

Single-repeated mode has a quirk worth knowing: when you post a single-repeated carousel to a platform with fewer allowed slides, the result still works because the image is the same — the platform just shows fewer text overlays.

Example: a 10-slide one-background carousel for Instagram + the auto-trimmed 4-slide version for X. The Instagram audience sees the full storytelling arc; X sees the opening panels — so front-load the hook and make the early panels carry weight.

This is not true for fresh-image-per-panel mode, where dropping slides means losing distinct content. One-background decks degrade more gracefully across platforms.

What AI is good at vs needs review

Per-slide carousels lean on AI image generation, which is strong at:

• Aesthetic backgrounds (warm-light coffee shop, moody coastal cliff, abstract botanical pattern)
• Generic scene illustrations matching your imagery style guide
• Cover-adjacent art (atmosphere of your book without showing characters AI might get wrong)

AI image generation is weak at:

• Specific characters with consistent appearance across slides (use Upload-Your-Own with your own art for this)
• Text inside the image — which is exactly why the tool never asks it to paint any; the panel editor owns the words
• Hands, eyes, fingers, complex architecture (regenerate, or swap engines via the per-post picker)

Single-repeated mode sidesteps most of these issues because the image is one canvas you control once.

Common mistakes

1. Picking 10 slides for a topic that needs 5. More slides isn't more value. Honest 5 beats padded 10 every time.

2. Generating per-slide art when single-repeated would be cleaner. A book quote carousel is almost always one-background. Five different background images for five quotes feels chaotic.

3. Forgetting per-platform caps. Your 10-slide Instagram carousel becomes a 4-slide X carousel. Check the Basics warning, decide whether to skip X for this batch or design for the trim.

4. Under-working panel 1. Panel 1 is the hook. Its text is what stops the scroll. Spend half your review time there.

5. Not regenerating panels that don't land. Each panel can regenerate alone. Use it — or swap engines via the per-post picker if a different model handles the aesthetic better.

What to do next

• For the basics of running a carousel-only campaign, carousel campaigns covers strategy.
• For when to swap engines per post, the per-post engine picker.
• For platform-specific options beyond carousels (Stories, Reels, first-comments), per-platform composer options.
• For the AI media providers themselves, Magnific (formerly Freepik), explained.