How to Create an AI Kiss Video: A Step-by-Step Guide

Making one of these clips is less about the tool and more about the two photos you feed it. The interface takes about ninety seconds to learn. Choosing images the model can work with is where the quality actually comes from.

Here is the whole process, in order, with the decisions that matter called out at each point.

What you need before you start

Very little, which is the point.

• A browser on any device. It works on an iPhone, an Android phone, a laptop or a desktop.
• Either two photos with one person in each, or one photo containing two people.
• JPG, PNG or WebP files.
• One payment for the video you decide to keep. One payment, one clip, no subscription and no sign-up.

You do not need an account, an email address or an editing app. There is more detail on the pricing page if you want to see exactly what a payment covers before you begin.

Step 1: choose two photos that will actually work

This is the step worth slowing down for. Almost every disappointing result traces back to the source images.

Look for photos where the face is large in the frame, roughly front-facing, evenly lit and in focus. A head-and-shoulders portrait is ideal. A full-body holiday photo where the face is a hundred pixels across is not, even if it is your favourite picture.

Then look at the pair together. Two portraits shot in similar light and from a similar height blend into one believable scene. A sunny outdoor selfie next to a dim indoor flash photo will produce a clip with a visible seam down the middle, because that is literally what the model is handed.

A few things to remove if you can: sunglasses, hats pulled low, heavy beauty filters, and anyone else in the frame. Crop group shots down to the one person you want. The full list, with the reasoning behind each item, is in the photo requirements guide.

Step 2: open the app and pick your mode

Go to the app. It loads straight into the create screen with no login wall.

There are two modes:

Two photos is the common case. You add one image per person and they are composed side by side into a single starting frame before anything is generated.

One photo with two people takes a single picture that already contains both faces. If you have a decent photo of the two of you, use this one. The lighting, camera and background already match, so the model has an easier job and the output usually looks more coherent.

Add your images and you will see a preview of the combined frame. That preview is exactly what the model receives, so if something looks wrong there: a face cut off at the edge, a strange crop, fix it now rather than after paying.

Step 3: choose a style and an aspect ratio

Two small choices with a large effect.

Style. Kiss is the default and the safest. French kiss asks for something more intimate and a longer moment of contact, and it is more demanding of your source photos. Hug moves shoulders and arms rather than just heads, which makes it the right pick for reunions, family clips and the younger-self trend. You can read what each one changes on the French kiss page and the hug page.

Aspect ratio. Pick 9:16 if the clip is going to Reels, TikTok or Stories. It fills a phone screen with no bars. Pick 4:3 if you want a more classic, photo-like frame, or if the video is going somewhere landscape-ish. The finished video inherits whichever shape you choose, so decide based on where you will post it rather than fixing it later.

Step 4: pay, then wait for the render

Generation costs one credit, and one credit is one payment: the amount is on the pricing page. Payment goes through Apple Pay, Google Pay or a card, and it is a single charge rather than a plan you have to remember to cancel. If you were given a promo code, there is a field for it before you pay.

Once the job starts, your video appears in the gallery with a pending status. Expect under a minute. You can close the tab and come back: the job keeps running on our side, not in your browser.

The variation in that wait is queue time rather than anything about your photos. When more jobs are running, yours waits longer before it starts. Nothing about the finished clip changes either way, so a slow render is not a sign that something has gone wrong with your images. If a job has been pending far longer than that, it is worth reloading the gallery before assuming it is stuck.

If the generation fails, the credit goes straight back to your balance and a retry button appears on that item. A failed run costs you time, not money.

Step 5: download, save and share

When the status flips to finished, open the video. You get a standard MP4, around five seconds long, that plays anywhere.

Download it to your device before doing anything else. From there you can post it directly, or drop it into whatever editor you normally use if you want to add music, a caption card or a second clip. Most people do not need to.

Two practical notes. Your gallery lives in the browser session you created it in, so if you switch devices you will not see old videos there. Keep the downloaded file. And if you plan to post it, add music in the destination app rather than baking it in, since the platforms give better reach to their own audio libraries.

When the result is not what you wanted

Work through this in order rather than paying for another run straight away.

• Faces look blended or one person vanished. There was more than one face in one of the uploads. Crop each photo to a single person and try again.
• The motion is stiff or barely moves. Usually a profile shot or a face that is too small. Front-facing and large in frame fixes most of these.
• The two halves do not look like the same scene. The lighting between your two photos is too different. Swap one out for a photo taken in similar conditions.
• Detail smears around the mouth. Some softening at the moment of contact is normal in current models. A higher-resolution source photo reduces it.
• The style is not what you pictured. Try a different style with the same photos before changing the photos. Kiss, French kiss and Hug produce genuinely different motion from identical inputs.

If you want to see the range of what good inputs produce, the examples page has finished clips, and the how it works page explains what happens between upload and download.

A short checklist

Before you press generate, confirm all six:

1. One person per photo, or one photo with exactly two people.
2. Faces large, in focus and roughly front-facing.
3. Similar lighting across both images.
4. No sunglasses, and no heavy filters.
5. The right aspect ratio for where you are posting.
6. Everyone in the photos has agreed to this, no minors, no strangers.

That last one is not a formality. Use your own photos and photos of people who said yes, and the rest of this is simply a nice thing to make for someone.