Scroll through enough AI-generated Instagram carousels and you'll start to recognize the pattern before you even read the text: a soft gradient background, a vaguely on-topic illustration repeated with small variations, the same three fonts every tool seems to default to. It's not that any single carousel looks bad. It's that they all start to look like the same carousel.
The usual explanation is typography or color choices. That's not actually it. The real problem sits one layer deeper: every slide gets the same visual treatment, no matter what that slide is trying to do. (This is the deep dive on visuals specifically; the complete guide to carousels covers hook, structure and consistency too.)
The shortcut most tools take
There are two common shortcuts. The first is generating one image per carousel, then cropping or repeating it across every slide. The second is pulling from a fixed library of generic icons tagged loosely by topic ("business," "growth," "health") and dropping one onto each slide. Both are cheap and fast. Both technically illustrate the topic of the carousel as a whole.
Neither one cares about what a specific slide is for.
Every slide has a different job
A carousel isn't one idea. It's a sequence of small, distinct beats: a hook, then a claim, then a piece of evidence, then a common objection, then a call to action. Each of those beats is doing different work, so each one needs a different kind of visual help.
The cover has to stop the scroll. That's where a strong, specific image earns its place, paired with a headline big enough to read in the feed thumbnail.
The slides in the middle have to be read. A point, a quote, a number, a comparison. Here a busy illustration usually gets in the way: text laid over artwork is harder to read on a phone. The reader came for the idea, not the decoration. What helps is structure. A numbered point that clearly says "this is point 3." A quote given room to breathe. One stat, big enough to land on its own. Two options side by side so the trade-off is visible at a glance.
The last slide has to get an action. That's another moment where a visual helps, because it signals "this is where you do something" instead of "here's another point."
Give all of those slides the same gradient-and-icon treatment and the visuals stop helping the reader track where they are. They become wallpaper. That's the sameness you notice in the feed.
What fixing it actually requires
Fixing this means the tool has to know what role each slide plays before it decides how that slide looks. The text gets planned first, slide by slide, with each slide tagged by its job: hook, point, proof, objection, comparison, call to action. Only then does the visual get chosen, based on that job rather than one template stretched across the set.
It also means accepting that more illustration isn't automatically better. The goal is the right visual on each slide, not the most artwork.
Where Kohapu lands on this
Every carousel Kohapu generates is planned slide by slide, with each slide's role decided before its design. The cover and a few key slides get an AI illustration matched to that carousel's idea. The content slides use typographic layouts chosen by what the slide is doing: a numbered layout for steps and mistakes, a quote layout for an insight or a story, a single-stat layout for a number, a split layout for a comparison. The visual system underneath (palette, illustration style, fonts) stays consistent with your brand, set once during onboarding, so every carousel looks like yours without looking like the last one. See how that fits into the full weekly batch and approval flow.