Two-person AI video remakes: assign each photo to the right role
Map two character photos to the left and right performers, check the upload slots, and review identity swaps before sharing a duo video remake.
Identify the roles before choosing the photos
Watch the reference once before uploading. In the APT. template, Photo 1 supplies the left performer and Photo 2 supplies the right performer. Read left and right from the viewer’s perspective in the reference, not from the performer’s own perspective. Use the slot labels together with the preview to decide where each person belongs.
Keep the role assignment fixed through the clip. If the camera moves or the performers cross, a person should continue the same role rather than inherit whoever happens to occupy the left side of a later frame. Note a distinguishing action or prop for each role so you can recognize it again during review.
Give each role one unambiguous character reference
Prepare a separate photo for each performer. A group photo can leave it unclear which face, clothing, and body belong together. Prefer one person per image, with the face unobstructed and enough of the body visible for the intended action. Use only photos you have permission to use for this purpose.
For example, if Alex should perform the left role and Sam the right role, keep an Alex-only photo ready for Photo 1 and a Sam-only photo for Photo 2. Compare their framing and lighting before you start. A tiny face in one image and a close-up in the other give you a less balanced starting point for judging the pair.
Fill the labeled slots and verify them before generating
Open the APT. template and add the first photo to Photo 1 · left, then the second to Photo 2 · right. Check both thumbnails after uploading. If a photo is in the wrong slot, remove it and upload the correct photo there; do not expect a filename to override the slot assignment.
This template already supplies its motion reference and fixed performance instructions. You do not need to upload that clip again or edit a hidden prompt. Prepare both photos, choose from the available resolution settings, and check the displayed output information and quote before you submit.
For your own reference video, write the mapping explicitly
The editable composer is a separate workflow for your own references. Add two photos and one video, then insert their actual reference labels using the material tags or @ menu. If the labels are @Image1, @Image2, and @Video1, a starting instruction is: “Use @Image1 for the left performer in @Video1 and @Image2 for the right performer. Keep those identities assigned to the same roles throughout.”
Then say what each reference contributes: the photos provide the people’s appearance and clothing, while the video guides movement, framing, and scene. Use the labels actually shown in your draft rather than copying numbers blindly. This example is for the editable composer; the fixed template already handles its reference instructions.
Review each person through crossings and occlusions
Watch the full result, not just its thumbnail. Follow the first person from the opening through the middle to the end, then repeat for the second person. Look for a change of face, hairstyle, or clothing after a turn, a crossing, or a hand passing over the face. Review both people again when they separate after an overlap.
Check movement and object contact separately from appearance. A recognizable face does not prove that a hand is holding the correct microphone or that a foot lands naturally. Keep a short checklist for identity, role, action, and props so a strong result in one area does not hide a failure in another.
Change the cause of the problem, then review again
If the people were assigned to the wrong roles from the start, recheck the upload slots first. If one person drifts later, keep the other photo and the template unchanged while trying a clearer, more compatible photo for the affected role. Changing both photos, the motion, and the instructions together makes the next result harder to interpret.
An explicit mapping is useful guidance, not a guarantee of exact identity, choreography, or an unchanged background in every frame. Decide whether the remaining issue makes the clip unsuitable before spending on another attempt. Only share the version you have reviewed, and handle final audio and captions separately when needed.