“A zombie couple hugs” names the outcome but leaves the model to invent everything that makes it legible. Who is who? What tells the viewer they recognize each other? Is the camera watching the decision or cutting away from it?
For a short AI zombie love story, write the recognition first. The apocalypse can be a quiet backdrop.
Use one clear photo per person in the AI Zombie generator. The first upload becomes Image 1, the second Image 2. Change the setting and sound to fit your own story; do not use photos without the subjects' permission.
Image 1 is the survivor in a dark jacket. Image 2 is the person they love, now subtly zombie-like but still recognizable from the reference photo. On a deserted street at blue hour, Image 1 sees Image 2, freezes, then slowly lowers their hand and takes one step closer. Image 2 responds with a familiar expression. They share a quiet embrace. One continuous medium shot with a restrained push-in. Preserve both faces, hairstyles, and clothing from their respective reference images. Muted gray makeup, no gore. Sound: wind through the empty street, soft footsteps, then a single relieved breath. No captions or text.This prompt assigns roles before describing motion. It gives the model a modest sequence—recognize, approach, embrace—rather than demanding a chase, a fight, a flashback, and a kiss in five seconds.
| What you see | Change in the next prompt |
|---|---|
| Faces merge or swap | Name Image 1 and Image 2 at the start; reduce camera movement and background characters. |
| The hug looks unnatural | Ask first for recognition and one step forward. Generate the embrace as a separate shot if needed. |
| Makeup hides the person | Say “subtle gray makeup, recognizable face”; remove descriptions of severe decay. |
| The story feels rushed | Use 10 or 15 seconds, or keep only the recognition beat in this clip. |
| Sound distracts | Specify one or two sounds; leave music and captions for the edit. |
Change one variable at a time. A longer prompt is not automatically more controllable, especially when it contains contradictory staging.
The warm-memory ending can be powerful because it sharply contrasts with the cold present. Make it as a second shot: the same two people alive in a specific, ordinary place—a kitchen, a bus stop, a small celebration. Then cut from a matched gesture, such as one hand reaching for the other, into that memory. Verify you have the right to use both source photos.
H3 Max reference-to-video creates one 5–15 second clip per request. Choose the join after both shots exist: match the gesture you actually got rather than forcing their opening frames together. The memory match-cut walkthrough gives you paired diner prompts and a concrete cut point. For the upload workflow, see the step-by-step AI zombie video guide.