AI hug video from two photos

The whole product is one sentence: you supply two photographs, and the video that comes back has those two faces in it. Nothing in the flow substitutes a generated person for someone you uploaded, and there is no library of pre-made couples you have to pick from instead. The hug style is one of fifteen the app can run, plus a custom prompt, and it is the one that animates an embrace rather than a kiss.

That sounds like the minimum a photo-to-video app should do, and it is. It is on this page because it is the thing people most often say they did not get. When we read 1,043 one- and two-star App Store reviews of other apps in this category on 13 September 2026, the sharpest cluster in the whole set was people describing an advert that promised exactly this and an app that would not do it.

These are verbatim quotes from public App Store reviews of other AI hug and kiss apps, read on 13 September 2026. They are quoted as complaints people made on a date, not as a claim about what any of those apps does today; every one of them may have changed since. They are here because they describe the same failure over and over, and it is the failure this page exists to answer.

The clearest one is a review of Boom AI, an app with 3,833 ratings of which 90% of the fifty most recent we read were one or two stars:

"I just saw one of their ads. It said you can take two photos and make them hug and kiss. This is false advertising. You can not use your own photos for this. The only option now is to write a text prompt and the app will generate the subjects for you. There is no way I can find to use your own photos."

A reviewer of a different app that ranks on the same search terms put it more simply: "this is the worst app ever because when I just put a picture in it, it's not letting me hug or kiss!" And a third, on the same app: "You have to pay for literally what they were showing in the video a.k.a. the AI hug trend."

The pattern in all three is the same. The advert shows two real photographs becoming one hug. The product turns out to be a prompt box, or a gallery of pre-made templates, or a paid tier the advert never mentioned. If you want the version the advert showed, that is what this page is describing, and there is a longer comparison of Boom AI and the alternatives if you are switching from it specifically.

The model that makes the video takes exactly one image as its starting point, so the first thing that happens is that your two photos are combined into one. That composition happens in your own browser, before anything is uploaded: each photo is fitted into half of a single canvas in the shape you chose, 4:3 or 9:16. The result is one picture with both people in it, which is what gets sent.

From there a video model animates that single frame into motion, and the direction it is given is the style you picked. Hug moves shoulders and arms and brings two bodies together; Kiss brings two faces together instead. The first moment of the clip, where the two halves of the composed image dissolve into one scene, is trimmed off before you get the file, because out of context it reads as a glitch.

This is also why the two photos matter more than anything else you can control. The model is not looking anyone up; it only has the pixels you gave it. The how it works page has the full sequence, and there is a walkthrough of the whole flow in making a kiss video from two photos.

Framing is the one thing that decides a hug. An embrace needs shoulders and ideally a little upper arm; a photo cropped tight to the face gives the model nowhere to put the arms, and the result is usually the awkward one people post as a joke. Head and shoulders, or head to mid-chest, is the shape that works.

Keep the two pictures consistent with each other. Similar brightness, a similar distance from the camera, similar sharpness. A daylight photo next to a dim indoor one looks like two pictures glued together regardless of how good the animation is. Old scanned prints are completely fine, and are often the entire point, but scan them flat, in focus, and without glare across the face.

Faces should point roughly at the camera with the eyes visible. Sunglasses, deep shadow across the eyes and strict side profiles all take detail away from the one thing the model needs most. The full checklist is in the photo requirements guide.

The table is the short version of everything above. It is also what the complaint cluster is asking for: the specifics that adverts leave out and that people only discover after paying.

It cannot invent detail that is not in your photograph. A blurry, low-resolution or badly scanned picture stays blurry in the clip, because there is nothing else for the model to work from. A strict side profile is the same problem: the far side of the face has to be invented, and invented faces are the ones that look wrong.

It will not hold three or more people straight. Group photos confuse who is embracing whom, so crop to two before uploading. Arms and hands are the hardest part of any hug to generate, so an odd finger or a strange sleeve turns up now and then even with good source photos.

And it makes no promise about how closely the result resembles you. Likeness is decided almost entirely by the photographs, which is a long enough subject to have its own page. What is promised is narrower and testable: the faces the clip is built from are the faces you uploaded, and a failed generation costs you nothing.