Zombie AI Video Prompt: What to Type for Each Beat

A zombie AI video prompt is short. Three or four lines, one per beat, describing what is in the frame rather than what the model should do with it. The templates below are what we use for the three scenes on the home page, and they're meant to be copied and edited rather than admired.

One thing before the prompts: the photos decide most of the result. A perfect prompt on a group photo fails; a plain prompt on a clear front-facing photo works. If you only have five minutes, spend them on the photo rules, not on wording.

The Rule: Describe State, Not Action

Models read this generation as a scene, not a script. Words like turning, transforming, becoming invite a moving camera and a mid-transformation frame — which is how you end up with grey skin in the daylight beat. Words that name a state ("already turned", "unchanged", "still") keep each beat still and separate.

So: name who is in the frame, what they look like, and where the light is. Nothing else.

Beat 1 — The Standoff

Two people facing each other in a dim room with cold overhead light.
The person on the left is already turned: pale grey skin, clouded
eyes, an ordinary jacket, no wounds. The person on the right is
unchanged and holding a gun lowered at their side. Both faces
clearly visible, front-facing, still camera, no blood.

Why it's written this way: the transformation is stated as a finished fact ("already turned"), the unchanged person is stated as unchanged, and still camera is doing real work — it's what keeps the beat from drifting before the cut.

Beat 2 — The Daylight Memory

The same two people in warm afternoon sunlight, outdoors, both
unchanged and healthy, standing a few steps apart and looking at
each other. Natural colours, soft light, no grey skin, no torn
clothing, gentle camera drift.

This is the beat that gets missed. If the daylight prompt doesn't explicitly say unchanged and no grey skin, the model carries the transformation across the cut and the clip reads as broken instead of sad. Copy that line in every time.

Beat 3 — The Lowered Gun

Back in the dim room, cold overhead light. The unchanged person
lowers the gun to their side and does not fire. No gunshot, no
wound, no blood. Both faces visible, still camera.

The negative statements at the end are not decoration. Without does not fire, a fair share of generations resolve the tension the obvious way — and a gunshot makes it a different genre, not a different ending.

Words That Break a Clip

• Blood, wound, gore, gunshot, bite, attack — they come back in the frame and push the clip into horror, which is not the format.
• Transforming, becoming, mutating — invites a mid-transformation frame, which is where the grey-skin leak comes from.
• Cinematic, 4K, masterpiece, ultra-detailed — the input photo sets the fidelity; these add noise to the prompt without changing the result.
• Anyone's name. Use "the person on the left". Names don't reach the model and can trip content filters.

Scene Variants

Swap the location line, keep everything else:

• Dim interior, cold overhead light — the default, most reliable.
• Wet city street at night — a wet city street at night, streetlights reflecting on the pavement; strongest contrast against the daylight beat.
• Open field at dusk — an open field at dusk, low warm sun; the pet version preset.

The Pet Version

Same structure, one line changed: a dog standing in a field at dusk, looking at the camera, calm for the first beat, and end on the two of you moving toward each other rather than the standoff. Keep anything aggressive out of the prompt entirely.

Exact timings and framing for each preset are on the template page.

What It Costs

There's no free tier and no subscription. You pay for the generations you actually run, which is why the queue moves at video-generation speed instead of being rationed behind a free allowance. One payment covers one video; length is 10 or 15 seconds at 720p, and packs are on the pricing page.

Output quality depends on the model and on your photos — the photography rules on the how-to page matter more than any prompt on this page, and they're the part you actually control. Anything outside them is a coin flip.

FAQ

Is it free? No — there's no free tier, and no subscription to cancel. You pay per video; the pricing page has the current numbers.

What if I don't like the result? The output depends on the model and on the photos you upload. We don't offer refunds based on how much you like the video, and we don't promise a specific likeness or a specific look — that's why the photo rules and the sample gallery come before payment. If a file is broken, the payment goes through without a video, or the export fails, that's a different case: write to support@zombieai.video and we'll sort it out.

Do I need to write a prompt at all? No. The three scenes on the home page run on presets. The prompts here are for when you want to change the location or the light.

Why doesn't my prompt make the faces more accurate? Faces come from the input photos, not the text. Change the photo, not the wording.

Is it a problem if I copy these word for word? No — that's what they're for. Thousands of clips will share these lines, and they'll still be different clips, because the photos are different.

Can I write it in another language? English works most reliably, because that's what the model was built on. Translate the scene line if you must, and keep the negatives ("no blood", "does not fire") in English.

Three lines, two photos, one clip. Start a generation.