Zombie Love Story AI Prompt: The Shot-by-Shot Template (Copy & Paste)

If you want to make the zombie couple video in a general AI video app, the prompt decides almost everything. Below is the structure that works with multi-reference models such as Seedance 2, plus the mistakes that make renders fail.

Don't want to write prompts? Zombie Couple AI Video builds this prompt for you from two photos and your choices.

The template

Upload two reference images in this order — the survivor first, the one who turns second — then use:

@image1 is a reference photo of a person. @image2 is a reference photo of another person.
Create a 15-second cinematic short film in four shots, photorealistic, shallow depth of
field, subtle film grain. The survivor is the person from @image1. The one who turned is
the person from @image2. Both keep the face, features and hairstyle from their reference
photo, clearly recognisable in every shot.
The person from @image2 has turned into a zombie: ashen grey skin, clouded pale eyes,
faint dark veins, torn dusty clothes. No blood, no wounds, no gore.
Shot 1 (0-3 s): inside a cold grey abandoned cabin, dust in pale window light. The survivor
raises a handgun with shaking hands, eyes full of tears.
Shot 2 (3-6 s): close-up of the zombie face; the zombie stops, tilts their head and slowly
recognises the survivor.
Shot 3 (6-10 s): the survivor lowers the gun, lets it fall and opens their arms wide.
The zombie runs and leaps toward them.
Shot 4 (10-15 s): exactly on the leap, hard cut to a warm memory: a sunlit golden wheat
field at golden hour, both alive and laughing, holding each other close. Warm light,
gentle slow motion.
Sound: wind, breathing, footsteps, then a soft piano score that swells at the cut.
No dialogue, no lyrics, no text, no subtitles, no watermark.

Why it's built this way

Part	Why it matters
@image1 / @image2 named up front	Multi-reference models otherwise mix up who is who.
Timed shots	Without times, models spend 10 seconds on the standoff and drop the memory.
"No blood, no wounds, no gore"	Real faces + gore is the most common reason renders get blocked.
"Hard cut" on the leap	The emotional payoff; soft transitions feel like a dream sequence instead.
No dialogue or lyrics	Avoids garbled speech and copyrighted-sounding music.

Variants

Pet version — replace the second reference with: "@image2 is a reference photo of a pet (same species, breed, fur colour and markings)" and the zombie look with "matted dusty fur, clouded pale eyes, a slow limping walk". End on "the person laughing and hugging the pet".

Family or friends — keep everything, but change the memory action to "laughing and hugging warmly like family" (or "like best friends") so the ending isn't romantic.

Other settings — change Shot 1's location: a snowy mountain road, a rain-soaked train platform, a ruined city street or a ruined chapel. Match the memory: fireplace, cherry blossoms, a rooftop at night or a wedding day.

Common failures

• Same person twice: two photos of one person produce a confusing clone.
• Group photos: the model picks the wrong face.
• Over-long prompts with lore: extra backstory steals time from the four shots.
• Asking for a famous song: often rejected; add a trending sound when posting instead.

Prefer to skip the trial and error? Our generator applies this exact structure, lets you choose the story and memory, and refunds failed renders automatically. See how to make a zombie couple video.