How-to
AI hug video prompts: five that look natural, with sample clips
Make an AI hug video from one or two photos, or write your own prompt. Seven rules from our own clips, five copyable prompts, sample clips and what to do when a run fails.
Published November 6, 2026
An AI hug video comes from a photo of two people, or two separate portraits, plus a short description of how they come together. You can skip the writing and pick a ready-made style, or write the motion yourself. Below are five prompts you can copy, plus the clips we made while working out what holds up.
Upload one photo of two people, or two portraits, then pick a style. No prompt needed.
The one-tap route: how The Hug works
The Hug takes either one photo with both people in it or two portraits. Two portraits are staged into a single frame first, and that frame is the opening shot of the video. Faces, hairstyles and clothing stay as they are in your photos.
Each style is a short scene, such as Reunion, Lift & Spin, The Standoff, One More, Walk Together or Princess Carry. Most come out in portrait; One More is landscape, and so is its sample. Clip length is 5, 10 or 15 seconds at 480p or 720p, and the free setting is 5 seconds at 480p. The pricing page lists credits for the rest.
Writing the prompt yourself
The effect carries its own motion text. For a hug the effect does not cover, use image-to-video in Studio with a start photo and your own words. The photo already holds the faces, clothes and light, so the prompt says what moves and when. The rules below come from the hug clips we made and kept, plus our dance clips, which we wrote the same way.
- Start before the hug. Distance, a turned back, side by side, or the moment one person spots the other all work as first frames. If the two are already embracing, the video can only continue it. A frame that sets up the first act lets the whole clip go to the turn and the release.
- Hold a stalemate with a repeating action. In The Standoff, one person stands with arms wide and the other refuses to look. Two seconds of statues reads as a frozen image. The clip works because he keeps renewing the invitation, so you can tell he has been standing there a while.
- Write verbs at full strength. Models discount motion, so "laughs" arrives as a twitch at the corner of the mouth. "Bursts out laughing, drops her arms and stomps over" comes out as a person.
- Give the clip a landing. A small closing action, such as a thump on the back after the embrace, turns "they hugged" into "she gave in".
- Prefer big movements. Two of our clips had fine hand work (a zip pull, a door handle) that rendered as vague reaching. Arms, steps and a dropped bag all read clearly.
- Lock the camera while people are moving, and push in slowly for the last beat.
- Let the start image set the shape. In our runs the video followed the frame it started from, so choose a portrait or landscape start photo on purpose.
A one-action clip such as a lift and spin worked straight from text, with no start photo. Scenes that depend on a specific starting pose, like two people facing off across a kitchen, needed a start frame to land.
Five prompts to copy
These are ours, written for a start photo of two adults at some distance from each other. Each follows the rules above. We have not run these exact prompts, so treat them as a first draft and adjust the beats.
[0-3s] She stands holding her arms out, looking at him. He pretends to study her, chin raised, not moving. Camera locked. [3-6s] She wiggles her fingers at him twice. He cracks first, grinning, and takes a step. [6-10s] He wraps both arms around her and lifts her half a step off the ground. She slaps his shoulder once, then holds on. Camera pushes in slowly. No talking. AUDIO: soft acoustic guitar, one laugh.
[0-3s] One person stands at a door with a hand on the frame, watching the other from a few steps away. The other looks up. [3-6s] They walk toward each other, slowing with each step. [6-10s] They stop, and the hug lands with both arms and a long exhale. One of them rests their chin on the other's shoulder. Camera locked, then a slow push-in. No talking. AUDIO: a quiet room, one sustained note.
[0-3s] The two stand a few steps apart, one of them bouncing on the spot. Camera locked. [3-6s] The bouncing one sprints in and jumps. The other catches them, both shouting. [6-10s] They stumble a half turn, still gripping each other, then lean back laughing. No talking. AUDIO: crowd noise, a whistle.
[0-3s] A parent steps in front of a graduate and straightens the collar of the gown. The graduate stands very still. [3-6s] The parent pats the graduate's chest twice, eyes down, holding something back. [6-10s] The graduate pulls the parent in with both arms. The parent's shoulders shake. Camera pushes in slowly. No talking. AUDIO: faint applause, a single piano note.
[0-3s] One person sits on a couch with their head down, elbows on knees. The other sits beside them and waits. Camera locked. [3-6s] The sitting one leans sideways until their shoulder rests against the other's. [6-10s] The other puts an arm around them and pulls them in, resting a cheek on top of their head. Both breathe out. Camera pushes in slowly. No talking. AUDIO: low strings, a slow exhale.
What goes wrong
When the staging step turns a pair of photos down, the card reads "The image engine turned this pair down. Try different photos. Your credits are back." A retry would send the same pair to the same image engine, so swap in different photos.
Your clip will look different from our samples, which use our own cast and the best take of each. The free setting is 5 seconds at 480p.
A skipped beat usually holds two main actions. Split it or cut one, because adjectives do not help. Our dance runs showed the same thing.
Questions
Do I need a prompt for The Hug effect?
Can I use one photo or two?
Will it keep my clothes and faces?
How long are the clips, and what does it cost?
Can I hug someone who is not in the same photo?
Pick a style with your own photos, or open Studio to write the motion yourself.