How-To Guide
How do you create TikTok slideshow images with AI?
Create TikTok slideshow images with AI in a repeatable order: brief each slide, generate one 9:16 frame at a time, and label the AI images before you post.
Start with the slide brief, not the prompt. Decide what each slide has to do, whether it hooks, proves, explains, pays off or asks, then write a prompt for that one frame: subject, framing, light, palette and the space where text will go. Generate at a vertical size the post can use, keep two or three options per slide, and compare them on a phone before you keep one. Export JPEG or PNG and check the frame: Shopify reports that TikTok recommends images of at least 1,080 by 1,920 pixels. TikTok's guidelines require a label on AI-generated or significantly edited content that shows realistic-looking scenes or people. Then post the images as a Photo Mode carousel, using the setup steps in our TikTok slideshow guide. No image model guarantees views, and an image you cannot source, label or repeat is a liability, not an asset.
- Brief each slide before you prompt: role, subject, framing, light, palette and where the text will sit.
- Generate two or three options per slide, then compare them at phone size; Genviral's checklist makes the same point.
- Export JPEG or PNG at a vertical size the post can use; Shopify reports TikTok recommends at least 1,080 by 1,920 pixels.
- Label realistic AI-generated or significantly edited content before posting; TikTok's Community Guidelines require it.
Brief the slides before you open a generator
Most weak AI slideshows fail before the generator runs. The images may be attractive one by one, but they do not form a sequence, so the post has nothing to pull a viewer forward. TikTok describes Photo Mode as a carousel format for photo content: carousel posts of still images that display one after another, which viewers can swipe through at their own pace. That description sets the production goal. You are building an ordered set in which each slide earns the next swipe.
| Slide role | The job of that slide | What the prompt has to fix |
|---|---|---|
| Hook | Why should someone stop scrolling? | One bold subject, recognisable before any detail, with high contrast so a headline can sit over it |
| Proof | What is the claim, and what makes it real? | The product, object, result or before and after, with an angle and light that show it honestly |
| Detail | What does the viewer need to know? | A close-up or a small set of steps, with a palette and space that leave room for app text |
| Payoff | What does the viewer get? | The finished result or the use in context, wide enough to read, in the same palette as the set |
| Ask | What happens next? | A clean frame with room for one closing line, save, send, try or visit the profile |
Genviral's guide to making slideshow images with AI opens its best-practice list with the first slide, calling the opening image the most important one, and asks for a consistent aesthetic across the series. Both are decisions you can make in a brief before a generator runs, and both are cheaper to fix on paper than in a batch of images.
Keep the brief to one page. Name the slide count, the role of each slide, the visual rule the whole set follows, and the one thing each frame must show. If you cannot say what a slide is for in a sentence, it will not survive the review in step five.
Original editorial diagram of the slide brief used in this article. The craft rules echo the published guidance summarised here; the role rows are our own planning structure.
Make it work for your brand
Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.
Write a prompt for one slide at a time
A single prompt for a whole slideshow produces a set of near-twins with no narrative. Write one prompt per frame instead, and keep the shared parts identical: the palette, the light direction and the level of realism. Change only the subject and framing that the slide brief calls for.
- Subject: name the object, person or scene, and the one detail that has to be visible. Genviral's example contrast is a generic city name against a specific landmark, dramatic lighting and realistic architecture, and the specific version gives the model something to hold.
- Framing: say vertical and state the crop. A frame built for 9:16 is tall, so name what sits at the top, middle and bottom instead of letting the model centre everything.
- Light: name the direction and quality, such as soft window light or hard side light, because flat light is what makes text hard to read later.
- Palette: lock two or three colours that match the rest of the set, so the slideshow reads as one series.
- Text-safe space: ask for an uncluttered band where the headline or caption will go. Genviral advises high contrast for text overlays, which is easier to design in the prompt than to patch afterwards.
- Negative: say what the frame must not contain, and keep typography out of it. Genviral warns that its named model struggles with narrow elements and small text.
Keep the prompt itself in a document with the slide number beside it. That log is what turns a lucky image into a repeatable set, and it is what you edit when one slide comes back wrong. OpenAI's image guide notes that both of its image APIs let callers adjust quality, size, format and compression, so prompt wording is not the only variable you control.
Generate several options and keep by rule
Generate variants of one frame, not a masterpiece on the first try. Genviral recommends two or three images at a time, adding that too many simultaneous generations can cause quality issues. OpenAI's image guide exposes an n parameter that returns multiple images from one request, which makes a small batch the default instead of an afterthought.
- Compare on the device the post will be watched on. A frame that looks detailed on a desktop monitor can turn to mush at phone size.
- Ask whether the hook slide still reads in about one second, from arm's length, before any text is added.
- Check the palette against the slide before and after it. If the neighbouring frames clash, one of them is wrong for the set.
- Keep the prompt and the export name for the chosen frame, so the next batch starts from a known setting instead of memory.
- Regenerate only the frame that failed. Replacing the whole set because of one weak slide costs time and breaks the consistency Genviral's checklist asks for.
One practical trap is size. OpenAI's recommended image sizes are 1024 by 1024, 1536 by 1024 landscape and 1024 by 1536 portrait, none of which is 9:16. Either request a vertical custom size where your tool allows one, or generate portrait and compose the frame on a 9:16 canvas, leaving deliberate colour above and below instead of a random crop.
Original diagram of the keep-by-rule loop. The batch size follows Genviral's two-to-three advice and OpenAI's n parameter for multiple images per request.
Get the size and file format right
The frame and the file are separate checks. TikTok's vertical canvas is not the same shape as an Instagram feed image, and the file limits differ too. Set these once in your export preset, then verify the first batch by hand.
| TikTok photo post | Instagram reuse | |
|---|---|---|
| Shape | 9:16 vertical; Shopify reports that TikTok recommends at least 1,080 by 1,920 pixels | Within a 4:5 to 1.91:1 ratio range, per Instagram's media reference |
| File | JPEG or PNG, the formats Descript names in its photo-compatibility note | JPEG, 8 MB maximum, per the same media reference |
| Width | Follow the 9:16 frame you compose | Minimum 320 pixels, scaled up if smaller; maximum 1,440 pixels, scaled down if larger |
| What changes when you reuse | A 4:5 image is wider than 9:16, so it leaves space above and below or loses its edges to a crop | The caption, cover and first slide should be rewritten for the different audience |
If you are reusing Instagram carousel art, put each 4:5 slide on a 9:16 canvas and fill the extra height with the brand colour. Crop only when the image still reads without its edges. Instagram's media reference sets the feed-image limits above; TikTok's frame is taller, so the Instagram file is a starting point, not the final export.
Two platform notes affect how the set ships. Shopify's walkthrough warns that Photo Mode may not be available on every account or in every region yet, so update the app and check before a campaign depends on it. Riverside notes that a post made with a template cannot be made swipeable. This article is about the swipeable Photo Mode set, not the auto-playing template route.
Original diagram of the format and disclosure gate. Each limit is attributed in the article to Shopify, Descript, Instagram's media reference, TikTok's Community Guidelines or the Commercial Music Library page, read November 2026.
Review every image at phone size
Review is where a generated batch becomes a usable set. Genviral's checklist ends with a phone test, so put the files on a phone before they go near a post. Read each frame at arm's length, then run the set in order, because a slide can pass alone and still fail in sequence.
- The first slide reads in about a second, before any overlay text is counted.
- Each frame carries one idea. If it needs a sentence of explanation, split it or cut it.
- No small text is baked into the image. Small generated type is the most common defect Genviral names, and it is easier to remove than to repair.
- The palette, light direction and level of realism match the neighbouring slides.
- No brand mark, product detail or human feature is distorted in a way that would mislead a viewer.
- The files are exported as JPEG or PNG, named in slide order, and kept with the prompt log.
A failed frame goes back to generation for that frame only. Keep the prompt log open, change one variable, and regenerate. Genviral's own troubleshooting advice is to regenerate with a different prompt, try a different reference image, and adjust expectations, which is a fair description of what iteration looks like.
Label AI images and keep the rights line clean
AI images carry obligations that the generator does not remove. TikTok's Community Guidelines require creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people, and warn that unlabeled content may be removed, restricted or labeled by TikTok's team. A slideshow of photorealistic scenes with a brand's logo counts as a place to be careful.
The same guidelines say you should only post content you created or have the right to share, and that unoriginal or reused material without anything new is ineligible for the For You feed. Do not repeat the claim that AI images avoid copyright issues. Image generation raises its own rights questions, and a label is a disclosure, not a licence.
If the same images go to Instagram, publishers have an optional is_ai_generated self-disclosure field in the media publishing request. The label and the rights check are separate tasks: one describes how the image was made, the other establishes that you may post it.
Sound is part of a photo post too. TikTok's Commercial Music Library page says businesses cannot use the general music library for commercial usage and should use the Commercial Music Library for organic content, ads and branded content. The library is described there as a pre-cleared collection of 1 million songs, free for businesses to use. Pick the track before you export, not after the post is drafted.
Superdirector, the site you are reading, can produce carousel images in your brand and hand them to you for posting. Its $199 Growth plan runs one AI-operated social account with standing posting permission granted at the account level, and publishes eligible Instagram carousels and single link-free X text posts. TikTok uploads stay with you. The $29 plan is MCP access and hands-on creation only, without an operated account. See how it works.