Zombie Love Story AI Video Prompt Guide - 2 Photos to a Short Film
Copy-paste prompts for making a zombie love story AI video from two photos. Four emotional beats, face consistency rules, and the match-cut trick that makes the embrace land.
A zombie love story AI video is a short horror-romance clip built from two reference photos: one person stays human, the other turns. The trend works because of one structural trick. A weapon comes up, comes down, and a hard cut lands on a warm memory of the same two people. Get the beats and the framing right and the clip reads as a real short film rather than a filter.
This guide gives you the prompts. Everything below is copy-paste ready for reference-to-video models such as Seedance or Kling. Replace {photoA} and {photoB} with your two images and swap the scene variables.
What makes the zombie love story AI video work
Three things separate a clip people rewatch from a clip that looks like a slideshow:
- Role clarity. The model has to know which face stays human and which one turns. If you leave that ambiguous, both faces drift and neither reads.
- A visible decision. Something has to change on screen. The gun comes up, then goes down. Without a decision there is no story.
- A matched cut. The final embrace of the cold scene and the opening embrace of the memory scene must line up in camera angle and body position. That alignment is the whole payoff.
Tell the story in four emotional beats
The cold scene runs ten seconds and carries four beats. Write them as an ordered list in the prompt, not as a paragraph. Models follow numbered sequences far more reliably.
Beat 1 - The weapon comes up
Character A aims a rifle at Character B, hands trembling, tears in eyes.
Keep the camera wide enough to see both bodies. A close-up here hides the relationship between the two characters.
Beat 2 - The detail that breaks them
Close-up: Character B's clouded eyes; a wedding ring still on their raised hand.
This is the turn. One prop carries the entire emotional load. A ring, a locket, a collar, a hospital bracelet. Pick one object and name it explicitly.
Beat 3 - The weapon drops
Character A lowers the rifle, lets it drop to the ground, opens arms wide.
"Opens arms wide" matters more than it looks. It sets the body position that the memory scene has to match.
Beat 4 - The approach
Character B runs toward the camera, arms reaching out.
Note that Character B runs toward the camera, not toward Character A. This gives you the hug from the front, which is the angle most likely to survive face-consistency drift.
The full cold-scene prompt
Cinematic horror-romance short film, {aspect} aspect ratio. Cold blue-grey color grade, {scene_cold}, ash and fog drifting in the air.
Character A (reference photo 1): a survivor. Keep the exact face, hairstyle and clothing from the photo.
Character B (reference photo 2): has turned into a zombie - milky clouded eyes, cracked grey skin, torn clothes - but keeps the same face shape, hairstyle and identity from the photo.
Story beats, in order:
1. Character A aims a rifle at Character B, hands trembling, tears in eyes.
2. Close-up: Character B's clouded eyes; a wedding ring still on their raised hand.
3. Character A lowers the rifle, lets it drop to the ground, opens arms wide.
4. Character B runs toward the camera, arms reaching out.
Keep both characters in the same screen positions across all beats, so the final embrace lines up for a match cut. Photorealistic, shallow depth of field, subtle handheld camera drift. Natural foley only. No dialogue, no text overlays, no watermark, no extra limbs, no face morphing between beats.
Keep both faces and roles consistent
Face drift is the number one failure mode. These five instructions cut it down more than anything else in this guide.
State the role in the first line. "Character A: a survivor" and "Character B: has turned into a zombie" appear before any story beat. The model reads role assignment as a constraint, not decoration.
Never let the reference be ambiguous. If your uploaded photos could plausibly be either role, the model guesses. Use the swap control so the person you want turned is the one the prompt calls a zombie.
Name the invariants explicitly. The phrase "keeps the same face shape, hairstyle and identity from the photo" is doing real work. Without it, zombie transformation drifts into a generic undead face and you lose the person entirely.
Lock screen positions. "Keep both characters in the same screen positions across all beats" prevents the model from swapping who stands on the left. Position swapping is what breaks the match cut later.
Say no to morphing. "No face morphing between beats" belongs in the prompt. It is cheap to add and it prevents the mid-shot identity slide that makes clips feel uncanny.
Match the embrace across the cut
The cut is the payoff. Two rules make it land.
Match camera angle, not just body position. The memory scene must state that framing matches the previous shot:
Framing must match the final embrace of the previous shot: same camera angle, same body positions, so the cut feels like one continuous hug across time.
Match the color grade deliberately. The cold scene is blue-grey with ash and fog. The memory scene is golden hour with sunlight and dust. The contrast IS the effect. Do not soften the cold grade to make the cut smoother. A hard tonal jump is what sells the time travel.
The full memory-scene prompt
Hard-cut match continuation, {aspect} aspect ratio. The same two people from the reference photos, now healthy and human, laughing mid-embrace in {scene_memory}, warm golden-hour color grade, sunlight and dust floating in the air.
Framing must match the final embrace of the previous shot: same camera angle, same body positions, so the cut feels like one continuous hug across time.
Slow motion, genuine laughter, hair and clothes moving in a warm wind. Photorealistic, soft lens glow. Warm emotional score. No dialogue, no text overlays, no watermark, no extra limbs.
When the weapon shot gets rejected
Content moderation on some video models rejects any frame containing a firearm. Rather than losing the generation, swap the prop. A trembling flashlight carries the same beat and passes every filter we have tested.
Cinematic horror-romance short film, {aspect} aspect ratio. Cold blue-grey color grade, {scene_cold}, ash and fog drifting in the air.
Character A (reference photo 1): a survivor. Keep the exact face, hairstyle and clothing from the photo. NO weapons of any kind in the scene.
Character B (reference photo 2): has turned into a zombie - milky clouded eyes, cracked grey skin, torn clothes - but keeps the same face shape, hairstyle and identity from the photo.
Story beats, in order:
1. Character A stands frozen, a trembling flashlight in hand, facing Character B.
2. Close-up: Character B's clouded eyes; a wedding ring still on their raised hand.
3. Character A drops the flashlight and opens arms wide.
4. Character B runs toward the camera, arms reaching out.
Keep both characters in the same screen positions across all beats, so the final embrace lines up for a match cut. Photorealistic, shallow depth of field, subtle handheld camera drift. Natural foley only. No dialogue, no text overlays, no watermark, no extra limbs, no face morphing between beats. No weapons.
Variant ideas
The same four-beat structure works with other subjects. Change the prop and the beat list, keep the framing rules.
Pet version
The pet must move as a real animal. Models default to giving animals human posture when the prompt does not forbid it, so the negative instruction is not optional here.
Character A (reference photo 1): the pet owner. Keep the exact face, hairstyle and clothing from the photo.
The pet (reference photo 2): has turned into a zombie - clouded eyes, dusty grey fur - but keeps its exact species, size, fur markings and identity from the photo. It must move as a real four-legged animal: natural trot, small jump fitting its size. NEVER give it human limbs, human posture or human gestures.
Story beats, in order:
1. Character A kneels, flashlight trembling, facing the pet.
2. Close-up: the pet's clouded eyes; its collar still on.
3. Character A drops the flashlight and opens arms wide.
4. The pet trots then leaps into Character A's arms.
Family version
Swap the wedding ring for a specific shared object: a house key, a childhood toy, a photograph. Anything that reads as a shared history in one glance.
Wedding version
Set the cold scene at a ruined venue and the memory scene at the ceremony itself. The ring beat becomes the centerpiece rather than a detail, so give it an extra second of screen time in the prompt.
Practical notes from testing
- Cold scenes are harder than memory scenes. Every warm, well-lit prompt we have run comes back clean on the first attempt. Cold, high-contrast sequences with props need more attempts. Budget for retries on the cold half.
- Aspect ratio changes the framing rules. In 9:16 vertical, "both characters in the same screen positions" needs the two bodies stacked in frame. In 16:9 they sit side by side. Write the aspect variable into the prompt rather than cropping afterward.
- Silence in the prompt means silence in the clip. We generate with "no dialogue, no text overlays" and add music in post. Generated speech over a two-face shot reads as uncanny far more often than it works.
- Keep the negative list short and consistent. Extra limbs, face morphing, watermark, text overlays. Longer negative lists make models drop earlier instructions.
FAQ
How many photos do I need for a zombie love story video?
Two. One for the character who stays human and one for the character who turns. Both should be front-facing with clear lighting. Sunglasses, heavy filters and extreme side angles all reduce face consistency.
Do I need a video editing app to do this?
No. The prompts above produce both clips, and the hard cut between them is a single splice. If you are generating through our generator, the two clips are assembled and the cut is placed for you.
Why does my character stop looking like themselves after the transformation?
The zombie instructions are winning over the identity instructions. Add "keeps the same face shape, hairstyle and identity from the photo" directly next to the zombie description and make sure you have not given the model two photos it could assign to either role.
Can I use a pet instead of a second person?
Yes, and it needs its own prompt. The pet version above adds a hard negative against human limbs and posture, which the standard prompt lacks. Without that line, models routinely give the animal human arms.
What if the model refuses to generate the weapon?
Swap the rifle for a flashlight. The refusal is almost always triggered by the prop rather than the scene, so replacing it keeps the beat intact. Restaging the shot without any prop at all also works, but the decision reads less clearly on screen.
Build it in the generator
Everything described here is already wired into the generator, including both prompt variants, the prop-swap fallback and the matched cut. Upload two photos, choose a scene, and the four beats run in order. You can also take the prompts above to any reference-to-video model you already use.