The eight most common fails in an AI duet video are merged faces, a face that drifts, one dancer who looks like a stranger, hands with the wrong number of fingers, steps that slide off the beat, a hallway that bends, outfits that change mid-clip and a stiff, frozen partner. On Jisoo Addicted AI there is no prompt box, so every fix comes from the same short list: change the photo, use two solo photos, or pick a higher model tier. A failed or blocked render returns its credits automatically.
Key takeaways
- Your levers are the photo, the number of photos and the tier. That list covers all eight fails below.
- A sharp, front-facing photo of each person fixes more than any other change, and it costs nothing.
- Two solo photos keep two faces apart better than one photo of both.
- Higher tiers hold faces, hands and fast motion more closely. At 720p, Standard costs 10 credits, Fast 17, Pro 20 and Ultra 37.
- The renders run on ByteDance Seedance models, which vary from run to run, so a second render of the same photos can differ.
- The site is independent and not affiliated with JISOO, ROSÉ, BLACKPINK, their agencies or labels, or TikTok.
To start over with better photos, open the generator. To understand why the duet behaves this way, read why a pasted prompt fails on the ADDICTED duet.
What should I check before blaming the render?
Check the photos first. A render copies the people in your photos into a 15-second hallway dance, so most visible problems trace back to how much of each person the model could see.
Three facts explain most of what follows. First, a duet needs two identities held apart for the whole clip, which is harder than a solo video. Second, the model works from your photos and the built-in dance reference, so the best correction is to improve what you feed it. Third, results vary between runs. If the photos are good and one render is flawed, a second render may be fine.
Keep one rule in mind while you read. Photos of celebrities, including JISOO and ROSÉ, and photos of minors cannot be uploaded. The tool is for you and people you know.
1. Why do the two faces merge into each other?
Faces merge when the model cannot tell the two people apart. It usually happens with a single photo that shows both people close together, small faces, similar lighting on both, or heads that overlap in the picture. During the dance the features slide between the two dancers, and you get siblings or one wobbling face.
The fix is to upload two solo photos, one face in each, instead of one photo of the pair. Each picture then states one identity clearly. If you only have a shared photo, pick one where both faces are large, sharp and fully visible, with no one partly behind the other. Moving up a tier also helps, because Pro and Ultra hold two identities apart more reliably than Standard.
2. Why does my face drift during the clip?
A face drifts when the photo gives the model too little to hold onto, so it fills in details and then changes them as the shot moves. The start looks like you and the end looks like a relative.
The fix is a better photo. Choose one that is sharp, front-facing or nearly so, evenly lit, and shows the whole face from forehead to chin. Avoid sunglasses, hats pulled low, heavy filters, strong side shadows and photos that were cropped from a group at low resolution. If the photo is already good, try Pro or Ultra for a closer hold on the face. Standard is a sensible first render, not the strongest one for faces.
3. Why does one dancer look like a stranger?
One dancer looks like a stranger when the model was given a clear face for one person and a weak one for the other, or when only one person was uploaded. If you upload a single person, the second dancer is generated as a passerby face, so a stranger is exactly what you asked for.
The fix is to upload a photo of the second person too. Use two solo photos with similar quality so neither face is the weaker one. If one person has a much clearer photo, the model tends to favor that identity. Matching the two photos for sharpness and angle gives you a more even pair. If you want the passerby on purpose, the single-person route is intended for that.
4. Why are there extra, missing or fused fingers?
Hands are the hardest part of any AI video and they get harder when two people touch. A hand crossing a partner's arm can pick up a finger, lose one or merge into the other person's hand.
You cannot instruct the model about fingers here, so the fix is on the inputs. Move up a tier, since Pro and Ultra render hands more cleanly than Standard. If a photo shows hands in the frame, check that they are clear and not clenched or hidden, because a messy hand in the photo is a bad starting point. A second render of the same settings is a reasonable cheap try, because this flaw is often random from run to run.
5. Why do the steps slip off the beat?
Steps slip off the beat when the motion in the render does not line up with the sound laid on top. The video is generated without sound and the trend's original audio is mixed in afterward, and the dance follows a reference built around that stretch of the song. Most clips line up well. A render can still drift when fast movement gets smoothed out or the model slows part of the move.
The fix is a higher tier, which keeps fast movement sharper, or a second render, since timing varies between runs. Check the result at full speed with sound on, not frame by frame, because small offsets that you would not notice in a post are easy to over-read when you scrub. The sound is the trend's original audio, and the song belongs to its rights holders.
6. Why does the hallway bend or change?
The corridor bends when the model loses track of the space, and corridors are among the hardest scenes because they are long, repetitive and full of perspective. Doors shift, the light strip curves, or the floor width changes between frames.
The Hallway Duet look keeps the original cream corridor with wooden doors and a warm light strip along the ceiling, and the setting is the same in every render. If the background still wobbles, render again or try Pro or Ultra for steadier surroundings.
7. Why does an outfit change mid-clip?
An outfit changes mid-clip when the model reinterprets clothing as the dancer turns. In the Hallway Duet look the dancers wear their own clothes, which come from your photos. If the photo shows only a face and a patch of collar, the model invents the rest, and the invention may shift.
The fix is a photo that shows the shoulders and upper body, with clothes that are clear and plain-edged. Busy patterns, thin stripes and small logos are hard to hold over 15 seconds. The clothes always follow your photos, so the way to control the outfit is to upload photos where you wear the one you want.
8. Why is my partner stiff or frozen?
A dancer looks stiff when the photo is posed in a way that fights the movement. A tilted head, a hand over the face, a crossed-arm pose or a photo taken from a steep angle makes natural dance motion harder for the model to produce, so it plays safe and barely moves that person.
The fix is a neutral, relaxed photo taken at eye level, with the arms down and the face visible. A candid photo works better than a formal pose for this reason. If one dancer is moving well and the other is not, replace only the weaker photo. If both are stiff, a higher tier gives the model more room to produce fluid motion.
What if the render fails or is blocked?
If a render fails or is blocked, the credits come back automatically. A failed attempt costs nothing. That is credits and not money: packs are non-refundable, and a render that finishes but looks wrong counts as a completed render.
That last point is why the order of fixes matters. Spend the free change first, the photo. Then change the cheap setting, which is two solo photos instead of one. Only then pay for a higher tier or a second render. At 720p, one Standard video is 10 credits and the Single pack is $6.99 for 10 credits. A Pro render is 20 credits and Ultra is 37, so move up only when the fault is the kind a tier addresses: faces, hands and fast motion.
Credits never expire, and there is no free tier. If you are not sure what a pack buys, the pricing page lists every option. The dances page shows the Hallway Duet look.
What is a good order of fixes?
A good order is photo, then photo count, then tier, then a second render. The first two cost no credits. The third costs more per render but often fixes the problem for good. The fourth is for random flaws only.
Here is how the eight fails map to the first fix to try:
| Fail | First fix to try |
|---|---|
| Faces merge | Two solo photos |
| Face drifts | Sharper, front-facing photo |
| One dancer is a stranger | Upload both people |
| Fingers wrong | Higher tier |
| Steps off beat | Higher tier or second render |
| Hallway bends | Second render, or a higher tier |
| Outfit changes | Photo with upper body and plain, clear clothes |
| Partner stiff | Neutral eye-level photo |
When the clip finally looks right, the next job is sharing it. How to post your ADDICTED duet on TikTok, Reels and Shorts covers the format, captions and music question.
Frequently asked questions
Should I change the photo or render again first?
Change the photo first when the problem is a face, a merged pair or a stiff partner, because the photo decides most of the result and costs no credits. Render again first only when the photos are already sharp and the flaw looks random, such as an odd finger in a single frame.
Do two solo photos really beat one photo of both of us?
For keeping two faces distinct, yes. Each solo photo states one identity, so nothing pulls the two toward a shared average. A single photo of both people still works if it is sharp and both faces are large and fully visible. Choose it when you have no separate pictures.
Which fails does a higher tier fix and which does it not?
Higher tiers help most with faces, hands and fast motion, because Pro and Ultra hold detail more closely than Standard. They do not rescue a blurry or sideways photo, so fix the photo first. At 720p, Pro costs 20 credits and Ultra 37, against 10 for Standard.
Is a stranger in the second spot always a mistake?
No. If you upload one person, the other dancer is generated as a passerby face, which is the intended behavior for that route. If you want a recognizable partner, upload a second photo of that person, ideally a solo photo with similar sharpness to the first.
Why does the hallway sometimes wobble in the background?
On Jisoo Addicted AI the hallway can wobble in a render because corridors are long, repetitive and full of perspective. The Hallway Duet look keeps the original cream corridor, so the setting is the same each time. If the background shifts, render again or move up to Pro or Ultra, which hold the surroundings more steadily.