Zombie AI Prompt for the Zombie Love Story Trend

The zombie love story is the AI short film that filled TikTok and Reels this October: someone you love has turned, you can't pull the trigger, and the last thing on screen is your happiest day together. Almost every version was made the same way, with an AI video model, two reference photos and one long zombie AI prompt.

Below is a prompt you can copy, written from scratch for this article, plus what we learned building this trend ourselves: which lines carry the story, what tends to break, and the setup that failed in most of our tests. If you would rather skip the prompt entirely, our Zombie AI Video template makes it from two photos.

What the zombie love story trend is

Every clip hits the same four beats, usually in 15 to 20 seconds:

1. The standoff. A dim cabin. One person aims a gun through tears while the person they love, now a zombie, staggers closer.
2. The choice. The gun shakes, drops to the floor, and the survivor opens their arms instead.
3. The leap. The zombie runs and jumps at them.
4. The match cut. Mid-air, the cabin cuts to a golden wheat field at sunset. The same jump lands as a hug from before it all happened, and the two of them spin, laughing.

The twist lives entirely in the last beat. The horror half is there to make the memory hit harder, which is why the whole video stands or falls on that cut.

The zombie AI prompt (copy and paste)

Upload two photos as reference images. The first is the survivor who holds the gun, the second is the one who turns. One person per photo, face clearly visible. Then paste this:

Vertical 9:16 live-action short film, 15 seconds, realistic and cinematic, hard cuts only. Two people: A is the person in reference image 1, B is the person in reference image 2. Match each face, hairstyle and skin tone to their photo in every shot, and ignore the photo backgrounds.

Part 1: an abandoned hunting cabin, late afternoon. Grey light through a dusty window, a bare plank floor, an overturned chair. A wears a worn olive jacket and has tears on their cheeks. A's skin stays clean and healthy in every shot, with no grey tint and no veins. B has turned into a zombie but is still clearly B: ashen grey skin, milky pale eyes, faint dark veins at the temples and neck, tangled hair, a faded red plaid shirt. The zombie look is make-up on smooth, unbroken skin: no cuts, scars, scratches, bite marks or wounds, and no blood or red stains on the face, mouth, hands or clothes.

Shot 1 (0-2s): close-up on A, face wet with tears, holding a pistol in both shaking hands, aimed just past the camera.
Shot 2 (2-4s): from behind A's shoulder, B shuffles in through the doorway and drags toward A.
Shot 3 (4-6s): tight close-up of B's zombie face, a slow blink, a quiet growl, more lost than frightening.
Shot 4 (6-8s): A breaks down, lowers the pistol and lets it fall to the floor, then spreads both arms wide and closes their eyes.
Shot 5 (8-10s): wide shot in exact side profile. A stands on the right of the frame with arms open, B on the left. B sprints and leaps at A. The shot ends with B in mid-air, a split second before they touch. No shot is fired.

Part 2: a memory from before, a golden wheat field at sunset, warm backlight and lens flare. B is alive and well, with healthy warm skin and bright eyes. Both wear light summer clothes.

Shot 6 (10-13s): match cut. Same side-on camera, same positions: B is in mid-air in exactly the pose the zombie had, and lands in A's arms. A catches B and they spin around, laughing, in slow motion.
Shot 7 (13-15s): close-up of A and B forehead to forehead among the wheat, eyes closed, smiling.

Sound: wind against the cabin, A's unsteady breathing, B's low growl and shuffling steps, the pistol clattering on the planks. At the cut, a soft piano melody rises over wind in the wheat and quiet laughter. No dialogue, no singing, no text on screen.

A few notes before you hit generate:

• Name the references the way your tool does. The prompt says "reference image 1" and "reference image 2". In Seedance you can write @image1 and @image2 instead; other tools have their own way to point at uploaded images, so check what yours expects.
• Clip too long for your tool? Split it at the match cut. Generate shots 1 to 5 as one clip and shots 6 and 7 as a second, with the same two photos, then join them in any editor. End the first clip on the mid-air frame and start the second on the same pose, and the cut still reads as one jump.
• Change the setting, keep the skeleton. A farmhouse kitchen, a car on an empty highway or a snowy rooftop all work. The shot order, the timings and the side-on leap are what make it this trend.

Why this prompt works

Numbered shots with timings

Give a video model one paragraph of story and it decides the pacing itself, which usually means a rushed gun drop and a memory that's over before it lands. A numbered shot list with seconds on each shot tells it what to show, in what order, and for how long. In our tests the model followed a list like this in order every single time.

Side-on positions for the leap

Shot 5 is wide and in profile, and it puts each person on a named side of the frame. That does two jobs: the jump reads clearly at phone size, and the next shot can repeat the exact same framing. Film the leap from behind the survivor and the zombie fills the screen, leaving nothing for the field shot to match.

The match cut is written into the prompt

Don't leave the transition to chance. Shot 6 says it continues the same leap, with the same camera and the same body positions, and that the person is mid-air exactly where the zombie was. Without that line, most models cut to an unrelated shot of two people in a field, and the payoff is gone.

The survivor's skin is described as clean

Models like to spread the zombie look around. Grey skin and veins creep onto the survivor's face, especially in shots where both people are in frame. One sentence saying the survivor stays healthy, with tears only, keeps the make-up on the right person.

It asks for no wounds or blood, in detail

"No gore" on its own isn't enough. In our first runs the model gave the zombie a scar across the forehead and dark red around the mouth, even with a prompt that called the zombie sad rather than scary. Listing what you don't want, cuts, scars, scratches, bite marks, blood on the mouth, hands or clothes, and calling the look make-up on unbroken skin cut that down a lot. Not to zero, though: faint red marks on the neck and around the eyes still showed up in some runs, so generate a few and keep the cleanest. It also keeps the clip clear of the graphic-content rules on TikTok and Instagram.

The zombie keeps their own face

The whole emotional hit depends on recognising the zombie as the same person who is laughing in the field. Saying that B is "still clearly B" under the make-up, and matching hair and skin tone to the photo, helps the model hold one identity across both halves of the film.

What we found in our own tests

We built this trend as a template on RapGlow, so we tried two very different ways of making it before choosing one.

A reference video plus face swap. Our first plan was to film the story once with AI actors, then swap each user's face in with a video-edit model. The person holding the gun swapped reliably in every test. The zombie did not. Across seven couple tests, the zombie wore the second person's face in only three, and the zombie and the wheat-field memory were both right in only two. In the rest, the original actor stayed on as the zombie, so the story suddenly had three people in it. Changing the photo, the facial expression or the gender pairing made no difference we could find. Pets were worse: in five pet runs the zombie never fully became the uploaded pet. Upload a cat, and a grey dog still did the leap while a cat showed up in the memory close-up.

Two reference photos plus a shot list. No source video at all, just the two photos and a prompt like the one above. This approach rescued the exact pair of photos that had failed every face-swap attempt. Across six runs, the right person or pet appeared in every role in five; in the sixth the zombie was only a partial likeness. All six kept the shot order and landed the match cut, and an orange cat stayed the same orange cat from the cabin to the wheat field, leap included.

The takeaway: if your tool offers both, write the story as a shot list with two reference images instead of face-swapping into an existing trend clip. It's also cleaner on rights, since you aren't reusing somebody else's video.

Pet variant: zombie dog or zombie cat

The same story works with an owner and a pet, and it's arguably even more of a tearjerker. Keep the prompt and change these parts:

• Reference image 2 is the pet. One animal, face and body visible, in good light.
• Describe the animal once and lock it in. Say what it is, breed and coat included ("a ginger tabby cat", "a golden retriever"), and that it's the same animal in every shot, including the leap. Otherwise the model can quietly swap in a generic dog for the action shot.
• Make the zombie version gentle. Cloudy eyes, dull greyish fur, a slow limp. Be extra firm about no blood: in our runs, pets picked up red around the mouth and eyes more often than people did.
• Rewrite the leap for the animal. The owner kneels with open arms; the pet bounds forward and jumps. After the cut, the pet leaps into the owner's arms in the wheat, the owner falls back laughing, and the last close-up is the pet nuzzling their face.

With reference photos, our tests held the right animal from start to finish: a cat, a golden retriever and a French bulldog each stayed themselves through the leap.

Photos that work best

• One person (or one pet) per photo. Group shots leave the model guessing whose face to use.
• Face toward the camera, in even light. No sunglasses, masks or hair across the eyes.
• Similar quality for both photos. A sharp photo next to a blurry one tends to give one character a much weaker likeness.
• Only people who said yes. Use photos of yourself and of adults who agreed to be in it. Celebrities and minors are off limits, and most video tools block them anyway.

Skip the prompt: Zombie AI Video

If you want the trend without the prompt, retries and editing, use the Zombie AI Video template. Upload two photos, the first holds the gun and the second turns, and you get a finished 9:16 vertical video with sound, ready for TikTok, Reels and Shorts. No subscription: it runs on one-time credit packs (see Pricing).

FAQ

What is a zombie AI prompt?

It's the text you give an AI video model to make the zombie love story trend: a shot-by-shot script for the cabin standoff, the dropped gun, the leap and the cut to a wheat-field memory. Most people pair it with two reference photos, one for each person.

Which AI video generator should I use?

Any model that accepts two reference images and a long prompt. We tested ours on Seedance 2.5 reference-to-video, where it kept both people and the match cut in almost every run. The prompt is written to be model-neutral, so it should carry over to tools like Veo or Kling, but expect to adjust how the references are named and how long a clip can be.

Why doesn't the zombie look like my second photo?

Usually the photo: a turned head, heavy shadow or a small, blurry face gives the model too little to hold on to under the make-up. Try a clearer, front-facing photo, and keep the line that says the zombie is still clearly that person. If you're face-swapping into an existing clip instead of using reference images, that's the setup that failed most in our tests.

How do I stop the AI from adding blood?

Spell it out. Say the zombie look is make-up on smooth, unbroken skin, and list what you don't want: cuts, scars, scratches, bite marks, blood or red stains on the face, mouth, hands and clothes. It still slips through now and then, so generate two or three and keep the cleanest.

Does the video come with sound?

Only if your model generates audio. The sound line in the prompt covers it: cabin wind, a growl, the gun hitting the floor, then piano in the field. If your tool makes silent clips, add a soft piano track in your editor and let it come in exactly at the cut.

Can I post it on TikTok or Instagram?

Yes, as long as everyone in it agreed. Turn on the AI-generated label, which TikTok, Instagram and YouTube all offer, and keep the clip free of blood so it isn't limited as graphic content.