FREDO BANG “OOUUH” UKULELE DUET · FIX GUIDE + GENERATOR
What You Want Me To Do AI Video: Fix the Face, Lip Sync, Hands and Blur
“What You Want Me To Do” AI video is what people call the Fredo Bang “Oouuh” ukulele duet trend; the phrase is a line from the hook, not the song’s title. The source is a 2018 phone clip shot in a small kitchen: Einer Bankz strums a ukulele on the left, Fredo Bang sings on the right and throws his head back on the long “ooh”. The AI version, which spread on TikTok in October 2026, puts two photos into those two places.
Broken versions tend to break the same few ways: a face that changes when the head goes back, a mouth that runs ahead of the song, fingers that melt into the ukulele, or a clip that looks sharp on your phone and soft on TikTok. This page goes symptom by symptom and says what to change, starting with the photos, which cause most of it. The generator below has the moves and the original sound built in, so the only things you control are the photos, the length, the model and the resolution.
2 photos (the same one twice for a solo) · 10 or 15 s on 5 models, 23 s on Seedance 2.5 · 9:16 · original sound · from 15 credits, 136 by default · failed videos refunded
Photo 1 · Ukulele player (left)
One person or pet per photo. The same photo in both slots makes a solo duet: two of you, one strums, one sings.
Photo 2 · Singer (right)
One person or pet per photo. The same photo in both slots makes a solo duet: two of you, one strums, one sings.
JPEG, PNG or WebP, up to 30 MB; each side 300–6000 px, at most 16 megapixels. One person or pet per photo, face sharp and facing the camera, no sunglasses. The player’s photo should show the hands and the outfit; the singer’s needs a clear front-on face. Use your own photos, your pets or adults who agreed.
480p · 9:16 · MP4 with the original sound
This video costs 136 credits. Charged when you start, refunded automatically if it fails.
About $6.80 of Studio credits per video. Credit packs are purchased separately.
See credit packs · Uses ordinary credits. Jean Phil trial credits are excluded.
Check your two photos before you pay
The two places need different photos. The ukulele player mostly needs visible hands and clothes; the singer needs a sharp, front-on face, because the singer faces the camera and throws the head back.
| Slot | What the model needs | What goes wrong without it |
|---|---|---|
| Photo 1 · ukulele player (left) | Hands and forearms visible, the outfit from chest to waist, the face at least three-quarters on | Made-up hands on the neck, fingers that bend the wrong way, clothes invented below the crop |
| Photo 2 · singer (right) | A sharp, front-on face with both eyes visible and nothing over the mouth | The face drifts on the head-back “ooh” and the eyes change shape |
| Both | One person or pet per photo, no sunglasses or beauty filter, the face filling at least a sixth of the frame | A group photo blends faces; a small face gets rebuilt from guesswork |
| The same photo twice (solo) | One good front-on photo, ideally down to the waist | A tight selfie makes both copies share the same invented body |
| A pet | The whole body, or at least the head, chest and front paws, at eye level | Human hands or a person-shaped body |
| Your own background | A vertical photo of an empty room, shot straight on at chest height, with nobody in it | Strangers in the shot, a tilted floor, a duet that seems to float |
Only have a low-resolution selfie from WhatsApp? Make it count:
- Ask for the original: WhatsApp compresses photos unless they are sent in HD or as a document.
- Crop to one person instead of uploading a group photo.
- Skip “enhance” apps that upscale with a beauty pass: they change the facial features the model has to keep.
- A video-call screenshot is the worst source; a fresh photo by a window beats it.
- Each side must be at least 300 px; below about 1,000 px, test with a 10-second Seedance 2.0 Mini video before paying for a long one.
Symptom, cause and what to change on the next try
Find what you see in your video, then change one thing at a time, so you know which change fixed it.
| What you see | Usual cause | What to change next time |
|---|---|---|
| The face doesn’t look like the person; the eyes look different | A small, soft or angled face; a filter; a cropped group photo | A sharp front-on photo; 720p; Seedance 2.5 or 2.0 Pro |
| Right face at the start, a different one by the hook | The thrown-back head hides the face and the model redraws it | Seedance 2.5 at 720p; a photo with the jawline clearly visible |
| Hair or clothes change | A head-and-shoulders photo leaves the rest open; a hat or hood in the photo | A photo that shows the outfit; take off what you don’t want kept |
| A cap you don’t own appears | The original player wore a cap; a photo cropped at the forehead leaves the top of the head unclear | A photo that shows the whole head and hair |
| Both people seem to sing | The singer went into Photo 1 | Swap the photos: the player’s mouth is set to stay closed |
| The mouth doesn’t match the song | The video is made silent and the original sound is laid on afterwards; the model ran slightly fast or slow | 23 s on Seedance 2.5; don’t add the TikTok sound on top; regenerate if it is off by more than half a second |
| Hands or ukulele warp; the strumming looks stiff | Fast strumming at 480p; hands hidden in the photo | 720p; Seedance 2.0 Pro or 2.5; a player photo that shows the hands |
| A pet gets human hands | Paws cropped out, or a drawing instead of a photo | A real photo with whole front paws |
| Sides swapped or faces mixed | Two look-alike people in similar clothes | Check the order (Photo 1 is left), different outfits, the Swap button |
| A third person shows up | Someone in your background or in one of the photos | An empty background; crop each photo to one person |
| Your background is ignored or bent | A cluttered, tilted or horizontal photo | A vertical, straight-on photo of an empty room |
| Rejected before it starts | Seedance 2.5’s review turns down celebrity photos, film stills and some other images | Credits come back; use your own photos, or a Seedance 2.0 model |
| Blurry once posted | A 480p video upscaled by the app, or a screen recording | 720p or 1080p; post the MP4 file itself |
Lip sync: why the mouth drifts
Here the video is made silent, following a depth map of the first 23 seconds of the 2018 take frame by frame. The original sound is then laid underneath, lined up with the first frame. The mouth lands on the words when the model keeps the timing of the moves; it drifts when the model speeds up or slows down a gesture.
The 23-second version on Seedance 2.5 follows the whole take, hook included, and drifts least. At 10 or 15 seconds the video stops mid-line; that is the cut, not a sync fault. The player never sings: if the player’s mouth moves, swap the photos.
If you add the official sound in TikTok afterwards, mute the video’s own sound first and start the sound at zero. Two tracks on top of each other make a correct video look out of sync.
Which model and resolution to try next
A finished video is charged even if you don’t like it. Before paying for the long version, test the photo on a short, cheap one, then step up.
| Model | Lengths | What it fixes best | Credits (15 s) |
|---|---|---|---|
| Seedance 2.0 Mini | 10, 15 s | Testing a photo before paying more | 20 at 480p, 39 at 720p |
| Seedance 2.0 Fast | 10, 15 s | A quick 720p once the face already works | 80 at 720p |
| Seedance 2.0 Pro | 10, 15 s | Hands and the ukulele; 1080p | 132 at 720p, 327 at 1080p |
| MiniMax H3 | 10, 15 s | A different take when Seedance struggles; no image review first | 89 at 720p |
| Seedance 2.5 (default) | 10, 15, 23 s | Faces through the hook; sync over the whole take | 90 at 480p; 23 s: 136 at 480p, 302 at 720p |
1080p doesn’t sharpen a soft photo; it only enlarges what the photo holds. Fix the photo first.
What each try costs: credits per video
The price depends on the model, the length and the resolution. One or two photos, the kitchen or your own room cost the same, and the still of the two of you is included. Dollar values are the Studio pack’s credit ($69.99 for 1,400 credits, about $0.05); packs are one-time purchases with no subscription, and credits don’t expire.
| Model | Length | Resolution | Credits | Studio value |
|---|---|---|---|---|
| Seedance 2.0 Mini | 10 s | 480p | 15 | ≈ $0.75 |
| Seedance 2.0 Mini | 10 s | 720p | 27 | ≈ $1.35 |
| Seedance 2.0 Mini | 15 s | 480p | 20 | ≈ $1.00 |
| Seedance 2.0 Mini | 15 s | 720p | 39 | ≈ $1.95 |
| Seedance 2.0 Fast | 10 s | 480p | 27 | ≈ $1.35 |
| Seedance 2.0 Fast | 10 s | 720p | 55 | ≈ $2.75 |
| Seedance 2.0 Fast | 15 s | 480p | 39 | ≈ $1.95 |
| Seedance 2.0 Fast | 15 s | 720p | 80 | ≈ $4.00 |
| Seedance 2.0 Pro | 10 s | 480p | 44 | ≈ $2.20 |
| Seedance 2.0 Pro | 10 s | 720p | 89 | ≈ $4.45 |
| Seedance 2.0 Pro | 10 s | 1080p | 219 | ≈ $10.95 |
| Seedance 2.0 Pro | 15 s | 480p | 64 | ≈ $3.20 |
| Seedance 2.0 Pro | 15 s | 720p | 132 | ≈ $6.60 |
| Seedance 2.0 Pro | 15 s | 1080p | 327 | ≈ $16.35 |
| Seedance 2.5 | 10 s | 480p | 61 | ≈ $3.05 |
| Seedance 2.5 | 10 s | 720p | 133 | ≈ $6.65 |
| Seedance 2.5 | 10 s | 1080p | 324 | ≈ $16.20 |
| Seedance 2.5 | 15 s | 480p | 90 | ≈ $4.50 |
| Seedance 2.5 | 15 s | 720p | 198 | ≈ $9.90 |
| Seedance 2.5 | 15 s | 1080p | 484 | ≈ $24.20 |
| Seedance 2.5 | 23 s | 480p | 136 | ≈ $6.80 |
| Seedance 2.5 | 23 s | 720p | 302 | ≈ $15.10 |
| Seedance 2.5 | 23 s | 1080p | 740 | ≈ $36.99 |
| MiniMax H3 | 10 s | 720p | 61 | ≈ $3.05 |
| MiniMax H3 | 15 s | 720p | 89 | ≈ $4.45 |
Default: Seedance 2.5, 23 s, 480p = 136 credits. The most expensive: 23 s at 1080p = 740 credits.
Sharp on your phone, blurry on TikTok
Blur after posting nearly always comes from how the video is uploaded, not from the original photo, even a DSLR portrait.
| Platform | Do | Avoid |
|---|---|---|
| TikTok | Post the MP4 file at 720p or 1080p; turn on high-quality uploads in settings | A screen recording; a 480p video that the app will upscale |
| Instagram Reels | Turn on “Upload at highest quality”; post the 9:16 MP4 | Running it through another editing app that compresses it again |
| YouTube Shorts | Upload the vertical MP4 and let HD processing finish | Judging quality in the first minutes, while it is still low resolution |
Refunds, retries and the watermark
| Situation | Credits | What you get |
|---|---|---|
| The generation fails or times out | Returned automatically | Nothing is charged |
| Seedance 2.5’s review turns down an image | Returned automatically | Change the photo or the model |
| The still of the two of you can’t be made | Returned automatically | Try sharper photos |
| The video finishes but you don’t like it | Charged | A new try is a new video; nothing retries on its own |
| Made before your first purchase | Normal price | A moving site mark; any purchase unlocks clean copies, past videos included |
| Sign-up credits | 3 credits | A video costs at least 15 credits |
Only ordinary credits pay for these videos; Jean Phil trial credits are excluded.
What a clean result looks like
Compare your video with these four on three points: does each face still match its photo after the head-back “ooh”, do the player’s hands stay on the strings, and does the player’s mouth stay closed? All four follow the same moves and the same track as the reference this generator uses.
Both faces stay the same from the first frame to the last; the player keeps both hands on the strings and his mouth closed while the singer does the chin hand, the grin and the swing.
A short cut: the cap, the chain and the plaid pattern hold steady, and the strumming hand stays on the strings through the opening lines.
The same photo in both slots: the dog strums with whole paws, no human fingers, and throws its head back on both hooks.
Two pets, two photos: each keeps its own fur, collar and markings, and the cat sings with its own muzzle on the full 23-second take.
Fixing a version made in another app
With a CapCut or PixVerse template you can’t pick the model or the resolution, so the photo is the only thing to fix, using the rules above.
With an app that works from a text prompt and no reference video, the moves are not the 2018 take’s and the music is often generated, so the mouth can’t match the real song. To get the real moves and the original sound you need a tool that follows the take itself, like the one on this page.
How to remake a What You Want Me To Do AI video without the glitches
- 1
Check both photos
Photo 1 shows the player’s hands and outfit, Photo 2 the singer’s sharp, front-on face; one person or pet per photo.
- 2
Test short
A 10-second 480p video on Seedance 2.0 Mini (15 credits) shows whether the faces hold.
- 3
Make the full version
Seedance 2.5, 23 s, at 480p (136 credits) or 720p (302) for the whole hook and the best sync.
- 4
Post the MP4 file
Download the MP4 and post it directly, with no screen recording and no second sound on top.
What You Want Me To Do AI video fix FAQ
The AI keeps changing my hairstyle and clothes. How do I keep my look?
There is no prompt to write here: the look comes only from your photo. Use one that shows the hair and the outfit you want to see, without a hat or hood unless you want to keep it. A head-and-shoulders photo leaves the model to invent the rest of the body.
My face looks weird and my eyes look different from the photo. How do I fix it?
Use a sharp, front-on photo with both eyes visible, no sunglasses or filter, and one person in it. Switch to 720p on Seedance 2.5 or 2.0 Pro. If the face is right at the start but changes on the hook, the thrown-back head is the cause; Seedance 2.5 handles it best.
The mouth doesn’t match the song. How do I improve the lip sync?
Choose 23 s on Seedance 2.5, which follows the whole 2018 take with its original sound. Check that the singer is in Photo 2 and don’t add a second track in TikTok. If it is off by more than half a second, generate again: results vary from one try to the next.
I used a DSLR portrait, so why does the video look blurry on TikTok?
The photo isn’t the problem: TikTok upscales a 480p video, and a screen recording makes it worse. Generate at 720p or 1080p, post the MP4 file itself and turn on high-quality uploads.
The hands look strange and the ukulele player doesn’t move naturally. What helps?
Fast strumming is hard for video models, especially at 480p. Choose 720p on Seedance 2.0 Pro or 2.5 and give the player a photo that shows the hands. A pet strums with whole paws on purpose.
All I have is a low-resolution WhatsApp selfie. Can I still get a decent video?
Yes, if each side is at least 300 px, but ask for the original first, crop to one person and test a 10-second Seedance 2.0 Mini video (15 credits) before the long version. Avoid “enhance” apps that upscale with a beauty pass: they change the face.
Remake your duet with the right photos
Two photos, no template to find and no prompt to write, with the original sound.
Add your photosMore trends with your photos on the home page.