AI Kiss Videos on Instagram: Best Practices

Instagram treats a five-second generated clip the same way it treats any short video: it decides in the first second whether to keep showing it to people. Everything below is about winning that second, and about not tripping over the platform's rules for AI content on the way.

If you have not made a clip yet, the app runs in a browser on the same phone you post from, which keeps the whole loop on one device.

Reels, Stories or Feed

Three surfaces, three jobs.

Reels is where a new AI kiss clip should go. It is the only surface built to show your video to people who do not follow you, and a short looping clip is exactly the shape it rewards. Loops matter here: a five-second video that plays three times reads as strong watch time.

Stories are for the second act. Post the finished clip to Reels, then use Stories for the source photographs, a poll between two versions, or a reply to the question everyone asked in the comments. Stories reach the people who already like you, so use them to deepen rather than to reach.

Feed works when the clip belongs to something larger: a carousel where the first slides are the original photos and the video is the payoff, or an announcement post where the clip is the illustration rather than the point. Feed rewards saves and shares, so it suits posts with some substance around them.

Why 9:16 is the default

Choose your aspect ratio before you generate, not after. The video inherits the shape of the frame the model starts from, so there is no lossless way to change it later.

For Reels and Stories, 9:16 is the answer. It fills the screen, leaves no letterboxing, and gives you vertical room to place text above the interface elements at the bottom.

Pick 4:3 when the clip is going into a Feed post next to photographs, or when you intend to drop it into a wider edit in another app. It is the more photographic-looking option, and it can be the better choice when the video sits inside a story rather than being the whole story.

Writing the caption

Assume the sound is off and the viewer is a stranger. The caption has to carry the context in one glance.

Lead with the specific fact. Not "we finally have a video together" but "four hundred days, two countries, no photo of us in the same room". Numbers, places and durations are what make a stranger care about two people they have never met.

Keep the first line short enough to survive the truncation. Everything after the fold is for the people who already stopped.

Then give people something to do. A question that is easy to answer gets more replies than a request to comment. Asking which of two clips to keep works better than asking people to tag a friend.

Length is worth a thought too. A caption of one or two lines suits a clip that explains itself. A longer one earns its space only when there is a story the video cannot tell on its own, and those are the posts people save. Decide which of the two you are writing before you start, rather than drifting into a paragraph because the box is there.

Hashtags, without the guesswork

Hashtags are a small lever now, not a growth strategy. Treat them as labels rather than as a lottery.

A workable approach: a few tags that describe what the video literally is, a few that describe the occasion or relationship, and one or two tied to your own account or series so the set is findable later. Put them in the caption or the first comment. It makes very little difference.

What does not work is thirty generic tags copied between posts. It signals nothing useful about the video, and it makes your posts look interchangeable to the people scrolling past them.

Disclosing AI, and why it helps

Instagram asks creators to disclose realistic AI-generated content, and it applies its own information label when its systems detect generated or heavily edited media. That label is coming whether or not you mention it.

Saying it yourself is better in every way. A short line in the caption is enough. It stops the top comment from being an accusation, it sets the frame as look at this edit we made rather than look at this footage, and it means the platform's label confirms what you already said instead of contradicting you.

The audiences for this format already know the technology exists. Being direct about it is not a weakness in the post. It is usually the thing that makes the comments pleasant.

Consent before you post

Two separate permissions are involved, and people often only get the first.

The first is permission to make the video. Use your own photos and photos of people who have agreed. No public figures, no strangers, and nobody under eighteen.

The second is permission to publish it, which is a different conversation. Someone can be delighted by a clip of their grandparents and still not want it on a public account. Ask specifically: is it fine if this goes on my Instagram. Family archive posts in particular deserve that question before they go anywhere.

What happens to the photos you upload is covered in the privacy and safety guide, which is worth reading before you upload someone else's face rather than after.

What to avoid

• Reposting a TikTok with the watermark on it. It is the single easiest way to have a Reel shown to nobody. Export a clean file and post it natively.
• Baking in music from an editor. Use Instagram's own audio library so the post can be found through the sound, and so a copyright claim does not take the whole thing down.
• Stacking filters on the clip. Generated video already has soft areas around the mouth and the edges. Extra grain and colour grading make those artefacts more visible, not less.
• Posting the identical file everywhere. Cut a separate version per platform, even if the differences are small. The TikTok guide covers what changes over there.
• Making a serious claim with a generated clip. These are for people who know the story. Do not let one imply an event that did not happen.
• Deleting a post an hour after publishing it. Reach on Reels builds over days rather than minutes, and a quiet first hour tells you almost nothing about how the post will end up doing.

A simple routine

Something like this holds up over a few weeks without becoming a job.

1. Collect and crop your source photos in one sitting. Single person per image, faces large, similar lighting.
2. Generate at 9:16. One clip is one payment with no plan attached, so testing a second style on the same photos is a small decision rather than a commitment. See the pricing page.
3. Post to Reels with a specific first line and an AI disclosure.
4. Follow up in Stories the same day with the source photos.
5. Reply to comments for the first hour.
6. Keep whichever framing performed and change one variable next time.

If you want a sense of what the finished clips look like at Reels size before you commit, the examples page has a set of them, and the app overview covers what running the whole thing on a phone actually involves.