Zombie Love Story AI Prompt & Photo Guide

Prepare two portraits and confirm permission

Begin with one adult per reference image. Choose a clear front view or a modest three-quarter angle with readable eyes, hairline and jaw. Soft, even light helps distinguish the features without introducing strong shadows that can look like part of the fictional transformation. Avoid sunglasses, beauty filters, motion blur and group photos when a clean individual portrait is available.

Obtain permission for both the transformation and the way you intend to share it. Someone who agrees to a private experiment may not want the result posted publicly. Confirm that every human participant is at least eighteen, and keep the planned scene non-graphic. A fictional setting does not remove the need to respect a person’s likeness or their preferred portrayal.

Photo A becomes the human survivor. Photo B becomes the infected partner in the opening and a healthy version of that same person in the final memory. Save those roles with the filenames. If you use a shared photograph, make two separate crops without stretching either face. Keep enough room around the head, and select another source if the crops become too small or blurred.

The studio accepts JPG, PNG and WebP up to ten megabytes, with a minimum readable dimension of two hundred pixels. These are technical input checks, not a quality score. Inspect the photographs yourself, then use the local previews to confirm the order before exporting the plan.

Choose the frame and write the four beats

Use 9:16 for the default vertical plan. Landscape and square options are also available, but changing the frame should not change who plays each role. Keep the pair comfortably within the image, with room around faces and shoulders for the embrace. A close crop can make identity clearer while leaving too little space for the physical contact, so consider both requirements.

The twelve-second target has four landmarks. At 0:00, establish the ruined greenhouse and the two adults. At 0:03, show the survivor lowering and setting aside an inert prop gun. It never fires, and there is no blood or injury. At 0:06, show empty hands and a gentle embrace. At 0:09, continue that pose in the selected healthy memory and end around 0:12.

Choose one final setting: a golden meadow, first snowfall or a rainy city. Avoid combining several memories in the same short ending. Write a brief closing line, but keep it as an editing note rather than asking the model to draw text into the image. Adding the caption afterward gives you control over spelling and placement.

The timestamps are a storyboard, not frame-level instructions guaranteed by every provider. Check the external model’s available durations and adapt the editing plan if necessary. Preserve the action order even when a particular tool needs a longer clip or separate shots.

See the timing table below for how the same four beats scale to 8, 10, 12, 15 and 25 seconds.

Copy the prompt and prepare a controlled render

The copyable prompt below is the default two-photo, meadow, 9:16 version generated by the same function used in the studio. It is included directly in this page’s HTML so you can read or copy it without opening a dialog. Customized selections in the creator produce corresponding variations. Keep a saved copy of the exact prompt submitted for each test.

Open an external video service that supports the type of identity references you need. Add the original approved portraits in the same A/B order, paste the instructions and choose a supported frame and duration. Read the provider’s current price and limitations before submission. A general text-to-video interface should not be assumed to preserve two people just because a prompt mentions their photographs.

Submit one controlled test and retain the task identifier. If the service reports processing or the connection times out, inspect that existing task before creating another. Repeated submissions can create repeated charges. The free studio does not hold your external account, monitor those jobs or promise a refund from another provider.

Do not change every variable after an imperfect result. First decide whether the problem concerns a reference, a role description, a pose or a cut. Change the smallest relevant element, save a new prompt version and compare the output against that specific correction. This keeps experimentation understandable and makes a budget limit easier to enforce.

Create a cinematic, non-graphic 12-second zombie love story with two adults in 9:16 format. Use reference image 1 as the human survivor and reference image 2 as the infected partner. Preserve both identities in every shot, including the healthy memory.

Four-beat storyboard:
0:00 Apocalypse: Establish the ruined greenhouse and the two adults. Keep the survivor on the left and the infected partner on the right.
0:03 The gun: Show the survivor lowering and setting aside an inert prop gun. The gun never fires; there is no blood, injury or threat toward the viewer.
0:06 The hug: The survivor opens their empty arms and the partner steps into a gentle embrace. Maintain the same camera angle and screen direction.
0:09 Memory: Cut on contact to a golden meadow at sunset, keeping the same embrace and composition. Both partners are healthy and fully human, and the story ends at 0:12.

Look: cool, desaturated ruins changing to warm, hopeful light. Subtle zombie styling, realistic clothing and restrained camera movement. Natural ambience, no copyrighted music. No gunfire, blood, bites, graphic injuries or violence. No logos or rendered text.

Editing note: add this closing caption afterwards: "I'd choose you. In every world."

Target timing is a creative plan, not a guarantee of exact model output. Review faces, hands and the transition before sharing. This is a storyboard and prompt, not a rendered video.

Plan the embrace match cut and finishing edit

The match cut links two shots through a similar pose and movement. Keep the survivor and partner on the same sides of the frame, maintain camera distance and height, and preserve which shoulder each head approaches. The setting and health of the characters change, but the contact should remain recognizable as a continuation of the same embrace.

In an editor, compare the final ruined-world frame with the opening memory frame. Find the strongest point of contact and try the cut there before adding a transition effect. A small trim can improve an otherwise compatible pair of shots. A long fade can make mismatched heads or arms more visible by showing both at once, so do not treat it as an automatic fix.

If the reference identities swap, or no frame contains the needed pose, the missing visual information may require a replacement shot. Editing can choose among existing frames; it cannot reliably restore a face that was never rendered correctly. Use the troubleshooting table to distinguish timing repairs from source problems before paying for another attempt.

Add your closing caption only after the images are approved. Listen for unexpected gunfire or other audio that contradicts the plan, and remove it. Use music you have permission to publish. Review the complete result both with sound and muted, because the visual story should remain understandable on its own.

Export, review and manage the source files

The studio’s download is a TXT prompt rather than an MP4. It contains your scene instructions and selected closing line, but no photo pixels. Keep the original portraits available for the external rendering step. Save your final video from that provider or your editor according to its download process, and retain an unedited copy of the source clips.

Before sharing, ask both participants to review their appearance and the intended audience. Check that the infected styling disappears in the healthy memory, that hands and shoulders make plausible contact and that no unexpected graphic content remains. Label realistic AI-made material using the destination platform’s applicable controls. TikTok’s own instructions require labeling AI-generated content containing realistic images, audio or video; consult the linked guidance when publishing.

In the current studio, selected photos are temporary browser previews. They are not sent to our server or saved to local storage. Reloading or using Start over clears that working session. Copying a prompt to another service introduces that provider’s privacy and retention practices, so read those separately. Clearing this page does not delete images that you uploaded elsewhere.

Store the final export, the prompt version and your permission notes together. If you later change the caption or crop, work from that project copy. The site currently has no paid video delivery, order tracking or media retention promise because video generation is not open.

How long should a zombie love story clip be?

Twelve seconds is the default because the story has four beats, and three seconds per beat is comfortable without feeling slow. Two seconds per beat is the practical floor. If your renderer caps the clip shorter or longer, keep all four beats and divide the total runtime into even quarters.

Two things stay fixed whatever duration you pick. The order of the actions never changes: the apocalypse, the lowered gun, the embrace, the memory. And the match cut always falls at three quarters of the runtime — the memory beat is the final quarter and it needs that room to land. If a service forces a different length, change only the total duration and the four timestamps in your prompt. Leave the action order, the safety lines and the A/B role assignment exactly as written; those are what keep both faces consistent across shots. Anything under eight seconds leaves too little room for the closing memory, which is the beat that carries the emotion.

A twenty-five second version is usually assembled from two shots that share a frame — the ruins and the memory — and joined at the cut, because most services cap a single generation well below that. Check the price and the per-generation limit before you pay for a second render.