What You Want Me To Do AI Video Prompt: the Fredo Bang "Oouuh" Duet

What is the "What You Want Me To Do" AI video?

It is an AI remake of one short phone clip. In 2018 the rapper Fredo Bang sang his song "Oouuh" in a small kitchen while Einer Bankz played a ukulele next to him; Einer Bankz later posted the take on TikTok, and in October 2026 people started swapping their own faces into it. The phrase people search, "what you want me to do", is a line from the hook. The song is called "Oouuh", which is why you will also see the trend called the Fredo Bang AI video or the Oouuh AI trend.

What makes a remake recognisable:

• Two people, fixed sides. The ukulele player stands on the left and keeps strumming; the singer stands on the right and does all the moves.
• The hook. About 14 seconds in, the singer tips the head back with eyes closed and holds a long "ooh". In the 23-second cut it comes twice.
• The room and the sound. A plain home kitchen, warm light, one continuous handheld shot, and the original recording underneath.

A good prompt has to carry all three. That is harder than it sounds.

Prompt or template: what each route can reproduce

Most "AI video prompt" answers assume the model can invent the scene. For this trend it has to copy one, so the route matters more than the wording.

So: if your tool can take a reference video, the prompts below get you close. If it can't, no wording will bring back the kitchen, the hook timing and the song, and two photos in the panel above is the shorter way.

Copy-ready What You Want Me To Do AI video prompts

Rename @Image1, @Image2 and @Video1 to whatever your tool calls its inputs. Keep the left/right lines: they stop the model from handing the ukulele to the singer.

Use @Video1 only for the two bodies' motion, their places in the frame, the timing and the slightly shaky phone camera. Ignore how the people in it look.

LEFT performer = the person in @Image1. They hold a small ukulele the whole time, one hand on the neck and one strumming, eyes mostly on the strings, mouth closed: they never sing.
RIGHT performer = the person in @Image2. They sing to the camera: hand up near the chin at the start, a grin and a laugh, then the head tipped back with the mouth rounded on a long "ooh".

Each person keeps the face, hair, skin tone, build and clothes from their own photo in every frame. Never swap or mix the two faces. If both photos show the same person, show two copies of them.

Setting: a small home kitchen with warm overhead light, a door behind them, wooden floor. One continuous take, no cuts, full body from head to feet, vertical 9:16. Exactly two people. Natural hands and fingers on the strings. No captions, no logos.

Vertical 9:16 phone video in a cosy home kitchen at night, warm ceiling light, one continuous slightly handheld shot, framed head to feet.

On the left, the man from @Image1 strums a small ukulele, smiling down at the strings and glancing up at his partner now and then. On the right, the woman from @Image2 sings to the camera with one hand near her chin, laughs, then throws her head back on a long held "ooh" and reaches both hands toward the lens.

Both keep their exact faces, hairstyles and clothes from their photos. Playful, affectionate mood, like a couple filming a duet for fun. No other people, no text.

Vertical 9:16 concert video. The person from @Image1 stands at a microphone on a small club stage under warm amber spotlights, a ukulele player on a stool beside them on the left. They sing a slow, catchy hook straight to the crowd, smile, then tip their head back with eyes closed on a long "ooh" while the crowd's phone lights sway in the dark.

Keep their face, hair and outfit exactly as in the photo. Medium-wide shot from the front row, slight handheld movement, no cuts. Realistic skin and hands, no captions or logos.

Use @Video1 for motion and timing only. Both performers are the same person from @Image1: on the left they strum a small ukulele with their mouth closed, on the right they sing to the camera and tip their head back on the long "ooh". Same face, hair and clothes on both sides, as if one person filmed a duet with a clone. Kitchen setting, one take, head-to-feet framing, vertical 9:16. No text.

Use @Video1 for motion and timing only. LEFT: the dog from @Image1 stands upright on its hind legs and holds a small ukulele in its front paws, strumming with a whole paw, never human fingers. RIGHT: the cat from @Image2 stands upright and "sings" with its own mouth, paw raised near its chin, then tilts its head back on the long note. Each animal keeps its own fur colour, markings and face. Real animals, no clothes, no human hands. Kitchen setting, one take, vertical 9:16.

swapped faces, the two faces blended, a third person, extra fingers, a guitar instead of a ukulele, a broken ukulele neck, the player singing, lips moving on the left, new hairstyle, different clothes, beautified face, cuts, zoom jumps, text, captions, watermark, logo

Beat sheet to paste under a prompt

Timed against the first 23 seconds of the original take. If your tool reads timestamps, append the rows that fall inside your length; they are also a quick way to check a result.

The 10-second cut ends before the hook, 15 seconds catches its start, and 23 seconds has it twice.

Keep your face, hair and clothes: lines and photo rules

Old, scanned or low-resolution photos work if the face is at least a few hundred pixels across; the upload limit here is 300–6000 px per side and 30 MB.

What the template costs per video

The panel charges credits per finished video, shown on the Generate button before you start. The price covers the still of both performers and the video; a failed or rejected job returns every credit.

Dollar figures use the Pro monthly plan (780 credits for $29), the lowest price per credit. "—" means the model doesn't offer that option: the full 23 seconds is Seedance 2.5 only, and MiniMax H3 renders 720p.

What You Want Me To Do AI video examples

Four versions of the same kitchen duet with the original sound: two pairs of friends, one dog playing both parts and a dog-and-cat duet.

Faces, the song and your photos

• Other people's faces. Use your own photo, your pet, or an adult who agreed. Celebrity photos and film stills are turned down by Seedance 2.5's review, which refunds the credits.
• The sound. The finished video keeps the original take's audio for personal posts. Buying credits is not a music license; a brand account or an ad needs licensed music.
• Your photos. They stay private inputs on your account and reach the video provider only through short-lived signed links. See the privacy policy.

Genjutsu AI is an independent tool, not affiliated with Fredo Bang, Einer Bankz, their labels, CapCut or any of the video models named here.

Writing prompts for other clips? The Genjutsu AI prompt guide covers reference-video prompts in general. Other two-person templates: the Wipe Me Down kitchen dance and the duo swap in the full editor.

Last updated: 2026-10-23