How to make the Hop Hop Hop AI video from two photos

Two photos, an eight-second clip, one hard cut. The order of the photos is the whole trick.

Updated 2026-11-02

This page is the practical version. If you only want to know what the trend is, read what is Hop Hop Hop AI instead. If you want the clip, read on.

Before you start: the two slots are not interchangeable

Every version of this trend is the same eight seconds. One person stands at the lens. They grin, point back at a friend far down a hallway, then open a flat palm above that friend's head and press it slowly down. The friend hops, then folds and sinks. At the hard cut they change places and it happens again, reversed.

Step 1 — pick the photo for each role

1. 1Photo 1 (the pointer): whoever you want in the big close-up first. A full-body or knees-up shot works, but the face has to be large enough in the frame to survive being blown up.
2. 2Photo 2 (the bouncer): whoever gets squished first. This one genuinely wants a full body, because the joke is a whole body sinking — if the legs are cropped the sink has nothing to compress.
3. 3Both photos should be the same kind of shot. A headshot next to a full body reads unevenly after the cut, because the model has to invent proportions for one of them.

Step 2 — pick a quality

720p is the native render: the clip is generated at that size and nothing else happens to it. 1080p is an AI upscale of that same 720p master, so it is sharper but it is not extra detail — it is a better-enough version for feeds that re-encode aggressively.

Both are 9:16 vertical MP4 with no watermark. The clip ships silent — see the note on sound below.

Step 3 — render, then check the two faces

A render usually takes a few minutes and you can close the tab; the job keeps running and lands in your creations when it finishes. When it arrives, check exactly two things before you post.

• Both close-ups are the right person. The person at the lens is the one who is large in frame, and that is the one most likely to drift, because the model is inventing a big face from a small crop.
• The far person actually sinks on the same beat as the palm coming down. If the palm finishes before the body moves, the causal read is gone and the clip looks like two people doing separate things.

Step 4 — check the sound is in time

The download already has the trend sound mixed under the picture, trimmed to the same eight seconds, so the hops land on the beat. You do not add anything. If you re-post the file somewhere that strips audio, re-add the sound in the app instead — the music belongs to its rights holders, not to this site.

When a render comes out wrong

Almost every bad render traces back to one of four causes, and three of them are fixable from the photo side without spending another credit on guesswork.

If a render fails outright — cancelled, interrupted, provider error — the credits come back automatically. You are only ever charged for a clip you can actually watch.