When an AI kiss result looks wrong: diagnose before you retry

First, find where the problem starts

Nearly every bad result traces back to one of three things: a source photo, the scene, or chance. Photo problems follow the person, so the same face looks wrong in every scene. Scene problems follow the setting, so a pair that worked elsewhere looks off only here. Chance shows up as a one-off that disappears when you run the same thing again. Telling these apart decides what to change.

• The same face is off in two different scenes: replace that person's photo.
• Only one scene gives trouble: try a nearby scene with a similar mood.
• One strange result after several good ones: a single retry is reasonable.

Symptom: one person does not look like themselves

This is the most common complaint, and it is almost always the photo. Look at the original upload for that person. A small face, a strong angle, sunglasses, a beauty filter or a deep shadow all strip out detail the generator needs. Swap in a straight-on, evenly lit, unfiltered shot and preview again as a still. Keep expectations realistic: a generated scene will not reproduce a face perfectly. The goal is that friends would recognize the person at a glance.

Symptom: faces collide, merge or hover apart

When noses overlap, lips blur together or the two heads float without touching, the head positions in your photos are usually fighting the pose the scene expects.

• Swap a dead-on stare for a photo with a slight head turn, which needs less rotation to meet the other face.
• Avoid photos where the head is tilted sharply up or down.
• Move between standing embraces and seated scenes, since the head positions differ between them.

Symptom: hands, arms or hair look wrong

Extra fingers, an arm bending the wrong way or hair melting into a shoulder are generation artifacts rather than photo problems. They tend to appear more where bodies overlap a lot, as in Bedside Lap Kiss or Bed Cuddle Kiss. If they show up in a still, rerun the still once. If they keep coming back, a scene with less overlap usually comes out cleaner.

Symptom: the clip drifts or the light looks pasted on

Sometimes the still looks good but a face shifts partway through the clip, or one person seems lit by a different sun. The usual cause is contrast between the two photos. A bright outdoor selfie paired with a dim indoor shot leaves the generator reconciling two lighting setups across every frame of a 5 to 7 second clip. Photos with similar brightness, or a scene whose light resembles both, reduce the effect. A clip that starts well and wobbles near the end can also be ordinary variation, so one rerun is fair before you change anything.

Stop rules that protect your credits

Repeating the same attempt and hoping for a different answer is where credits disappear. A few simple rules keep that in check.

• Never test a new photo pair with a video run; preview it as a 5-credit still first.
• Two misses in a row with the same inputs means changing something before a third attempt.
• Change one thing per attempt, the photo or the scene, never both at once.
• If you see an insufficient-credits message, check your balance and credit history on /profile before trying again.
• Download good results right away. They are deleted after 7 days, and recreating them costs credits again.