TikTok slideshows are the format where a carousel of still images scrolls through with text overlays and a music track. They're one of the highest reach formats on the platform right now, especially for accounts with fewer than 10,000 followers. They cost less to produce than video, ship faster, and the algorithm serves them aggressively when the hook lands.
The problem has always been the middle step. Turning an idea into ten coherent, on-brand slides with consistent typography, good imagery, and a narrative arc used to take an afternoon in Canva. With AI tools in 2026, it takes minutes.
Here's the workflow we use at Supapost, the exact one we built our product around.
Why slideshows beat video for growing accounts
A few things working in the format's favour.
Slideshows are faster to produce. You can ship five a week without burning out. The algorithm rewards posting cadence more than polish for accounts under the follower threshold, and cadence is where slideshows win.
Text-led hooks outperform video hooks. A viewer can read the first slide in about half a second, decide to stick around, and keep scrolling. Video hooks take three to five seconds before the viewer commits. That's a much wider margin for error.
They repurpose trivially. The same deck cross-posts to Instagram carousels with zero changes. Video needs re-editing, new captions, and often a different aspect ratio.
None of this means video is dead. But if you're choosing where to put the first hour of your day as a creator, slideshows have a better expected return until your account passes the point where video reach catches up.
The four ingredients of a good slide
Before the AI step, a slide needs four things working together.
First, the hook. Slide 1 does about 90% of the work. Either the text or the image has to stop the scroll. Both is better.
Second, consistent typography. One heading font, one body font, both big enough to read on a six-inch screen at arm's length. Serifs under 32pt don't render.
Third, an image that earns its pixels. Either illustrative (drives the idea forward) or atmospheric (sets the mood). Stock photos with a watermark or a model smiling at the camera don't count.
Fourth, a payoff on the last slide. A CTA, a resolution, a reveal. Viewers who swipe all the way through are the ones most likely to convert, so reward them for the effort.
Most AI-generated slides fail on the image step. They produce a generic "professional looking" stock photo that tells the viewer nothing and pattern-matches to spam. Avoid that.
Step 1: prompt the deck structure
Write ten hooks before you touch a design tool. Do this in a notebook or a plain text file. Don't let the AI do it blind. AI tends to write flat, listicle-style hooks unless you give it voice.
Here's an example prompt I'd use:
Write ten hooks for a TikTok slideshow aimed at new parents. Topic: cheap healthy dinners. Tone: friendly, slightly self-deprecating, no influencer cadence. Format: each hook is one sentence, reads like something a sleep-deprived mum would say out loud.
Pick the three hooks that feel true and build one deck around each. Don't pick the cleverest. Pick the one you'd actually say to a friend.
For each hook, sketch the arc in ten lines.
- Slide 1: hook
- Slide 2: promise or stakes
- Slides 3 to 8: the meat, one idea per slide
- Slide 9: reframe or twist
- Slide 10: payoff and CTA
Step 2: generate the imagery
This is where AI earns its keep. Modern image models like FLUX Pro, Google Nano Banana Pro, and Higgsfield Soul produce imagery that's a whole tier above what was possible eighteen months ago. But they still need steering.
Two rules for slideshow imagery.
Lock an aesthetic before you generate. Pick the visual language before slide 1 and stay in it. That means a consistent colour palette, consistent framing (closeup vs wide), and consistent subject treatment (illustrated vs photo vs 3D render). Mixing destroys the flow.
Generate images that serve the text, not the other way round. Decide what each slide says first, then generate imagery to support it. If you generate images first and try to caption them, the text will feel bolted on.
If you're building a character-led series (say, a recurring host or AI influencer), generate the character once and reuse their identity on every future slide. Supapost handles this automatically with an identity anchor. You generate a persona, pick the best image, and every subsequent generation references it so the face stays consistent across scenes. That's the difference between "who's that person?" and "oh, it's Mary again."
Aspect ratio: 1080 by 1920 pixels (9:16). TikTok letterboxes anything else. Build for that from the first pixel.
Step 3: layer the text
Typography is where most AI-generated decks die. Three defaults that work.
- Heading: 80 to 120pt, tight tracking (around minus 2%), 1.1 line height
- Body: 36 to 48pt, 1.3 line height
- Caption or source: 24pt, 0.7 opacity
Pick one heading font and one body font. Space Grotesk or Inter for modern, DM Serif or Playfair for editorial, Oswald for punchy. Don't mix more than two faces.
One rule on text backgrounds. If you put text over a photo, add a 60 to 80% black rectangle behind the text block. The contrast drop kills engagement. Every slideshow you've ever seen that felt "professional" had that rectangle.
Step 4: export and publish
TikTok accepts JPG and PNG. JPG keeps file sizes small. Any CDN will happily serve 200 to 400KB per slide if you compress sensibly. Don't over-optimise, just don't upload 8MB masters.
For publishing, you have two options. You can do it manually, exporting ten JPGs, opening the TikTok app, and uploading as a carousel. Or you can automate it. TikTok's Content Posting API lets approved developers push slideshows straight to the user's draft queue, where you review in the TikTok app and hit publish. Much faster if you're shipping a deck a day.
The one-hour workflow
Here's what our power users run, end to end.
- Ten minutes: pick topic, draft ten hooks by hand, pick the three best.
- Fifteen minutes: generate imagery. One influencer or one visual theme. Lock the aesthetic.
- Twenty minutes: lay out text over each slide. Review the arc. Does it build?
- Ten minutes: tune typography. Make the hook slide 20% louder than you think it should be.
- Five minutes: schedule to TikTok, or publish directly to drafts for mobile review.
Under an hour, ship quality. You'll get faster as the system beds in.
Where Supapost fits
Our product is this workflow as a single web app. You describe the deck you want in one prompt. Something like "ten slides about cheap healthy dinners for new parents, warm palette, self-deprecating tone". Our AI agent drafts the slides with proper typography and positioning, and you tweak what needs tweaking rather than starting from a blank canvas. The identity-anchor feature keeps AI characters consistent across scenes. Scheduled publishing ships posts to TikTok through the official Content Posting API.
We built it because we were running this workflow manually every day and got tired of the tool juggling. It's a monthly subscription, no contracts, cancel anytime. Start free at supapo.st.
What comes next
Once you've run this workflow ten times, your posting cadence becomes the main lever. Most creators under-post. Commit to one deck a day for thirty days, measure which hooks land, and double down on those angles. That's the compounding loop.
If you want the agent to handle the whole thing end to end, from deck generation to cover art to scheduled publishing, we've made that available too through our MCP server at mcp.supapo.st. Claude Code, Cursor, OpenCode, OpenClaw, and every other MCP-capable agent can drive it. Install the skill with npx skills add supapost-dev/skills and your agent knows the workflow above better than most humans.
Either way, pick a topic, write ten hooks, and start shipping.