Can I upload one image twice?
That does not provide two distinct identities. Use separate consenting adult portraits, or use fictional characters if you only want to explore the plan.
two photos zombie video
A two photos zombie video workflow assigns one portrait to the survivor and the other to the infected partner. This page explains how to prepare those references before using the free storyboard and an external renderer.
Last updated: 2026-10-15
Apply the A/B rule before you touch the prompt
Photo A is the identity that remains human through the ruined-world sequence. Photo B is the identity that receives subtle infected styling. In the final memory, both are healthy versions of those same people. This order is a production convention, not a gender rule. Decide it with the participants and label the source files so that you can repeat the choice in another interface.
Open both photos side by side before loading them into the studio. Confirm that each image contains one adult, the faces are recognizable and you have permission for the intended fictional transformation. If you reverse the roles later, use the swap button and save a new prompt version. Do not silently exchange the reference files while keeping notes that describe the old order.
Some external services number their image inputs differently or treat the first attachment as a scene reference. Read the chosen tool’s instructions. The A/B labels in this guide explain your intended roles; they cannot force a provider to interpret unsupported reference fields. A small controlled test is safer than assuming every image-to-video interface handles identity the same way.
Choose light, angle and detail that explain a face
Prefer a clear front view or modest three-quarter angle, with both eyes visible and the head comfortably inside the crop. Soft daylight usually makes facial features easier to read than a strong colored light or a silhouette. You do not need identical lighting across the pair, but avoid giving one identity much less usable information than the other.
Inspect the original image at a useful size. Look for the hairline, eye shape, nose, mouth and jaw without relying on a beauty filter. Motion blur, extreme smoothing and heavy compression can remove the details that distinguish one person from another. Upscaling a tiny crop may make the file larger without recovering that lost information. Choose another source when the face is too small.
The browser accepts JPG, PNG and WebP files up to ten megabytes and rejects unreadable images or dimensions below two hundred pixels. Passing these checks only means the file can be used for a preview. It does not mean the face is sharp enough, the person is an adult or the resulting video will preserve the likeness. Those require separate review.
Avoid six reference mistakes that create ambiguity
First, avoid group images in a slot intended for one person. Several faces make the intended identity ambiguous. Second, avoid sunglasses or objects that cover the eyes, because they remove useful identity information. Third, replace strongly filtered images whose facial proportions no longer match the person. Fourth, avoid tiny face crops expanded to a much larger canvas. Fifth, do not pair an extreme profile with a straight-on portrait unless you have tested that combination. Sixth, do not upload the same shared couple image into both roles and expect the model to infer the mapping.
Fix the input itself before adding a long list of corrective adjectives. For example, a clearer replacement portrait is usually a more direct response to an obscured face than repeatedly insisting on perfect likeness. Keep descriptions compatible with the photographs: instructions about different hair, age or facial features can compete with the identity reference.
Make one change per test so you can learn which correction helped. Record the two filenames and the order used for each external task. If all inputs and the prompt change together, comparing outputs becomes much less useful.
Crop a shared picture without changing either person
When a shared photo is your best source, make two independent crops. Keep each head upright and preserve the original width-to-height proportions. Include a little space around the hair and shoulders rather than cutting directly through facial features. Exclude the other person’s face from the crop when possible, but do not distort the selected face to make the framing fit.
Compare the two resulting crops at approximately the same displayed size. If one face is much smaller or more compressed, search for another source. A shared background does not compensate for missing identity detail. Avoid using a heavily edited wedding or event image if the retouching has changed the recognizable features that the participants want preserved.
Save the original and both crops separately. The free studio only previews the files you select, so it will not create or store these crops for you. When you later use a video service, upload the same approved versions in the same A/B order. If that service alters the crop automatically, review its preview before submitting the chargeable task.
Test continuity across both worlds, not just the first frame
A convincing opening is only the first check. Compare the survivor at 0:00 and 0:03, then inspect both faces during the 0:06 hug and 0:09 healthy memory. The infected partner should lose fictional zombie styling in the final scene while retaining recognizable identity. A beautiful ending with a different person is an identity error, not a successful transformation.
If the roles swap, confirm the actual submitted attachments and the wording that refers to them. If they are correct, simplify camera movement and physical action for the next test. Keeping the pair on the same sides of the frame makes the intended continuity easier to specify. Avoid adding an orbit, spin or sudden perspective change while troubleshooting.
The downloaded prompt contains role instructions but no photo pixels. Keep your originals available for the external provider. Delete local previews with the remove controls or Start over when you finish. If you uploaded the images elsewhere, use that provider’s deletion process as needed; clearing this browser does not delete files stored by another service.
Frequently asked questions
Do the portraits need matching backgrounds?
No. Face visibility and consistent identity information matter more. The prompt supplies the ruined setting and the final memory environment.
Does a larger file always work better?
No. A large file can still contain a blurry or heavily filtered face. Inspect facial detail rather than judging only by file size.
Can I use a cropped group photo?
Only if each crop clearly isolates one consenting adult and retains usable detail. Choose another source if surrounding faces remain confusing.
Why are photos absent from the TXT file?
The download is a written plan. Your photos stay in the local preview and must be attached separately if you use an external rendering service.
For realistic AI posts, see TikTok’s official AI labeling guidance. Check the current rules when publishing.