Piano bars: diegetic keys on a locked Yamaha, not a generated crooner

Vidu Q3 from the piano still. A talking singer is a likeness. Meter the room mix before you post.

A piano-bar clip that books is the instrument in this room: lid, badge, bench, the lamp that actually sits on the fallboard. Motion is keys and a camera that does not invent a second grand. A generated crooner is a person you did not hire and a voice you do not clear. Animate the Yamaha. Leave the mouth off the file. Measure the mix you got.

/for/cocktail-bars is the drink-build page: shake, strain, garnish. /for/musicians is the performer. This post is the room's piano as the SKU. A cocktail reel is not a locked grand. A musician reel is not the house Yamaha.

Lock the Yamaha. Then add keys.

Photograph the piano where it sits — same house left, same brick, same stool if the stool is part of the listing. That still is the contract. A paragraph that says "moody jazz bar, singer at a black grand, golden hour" invents a cousin room and a face. Image-to-video keeps frame one as the file you sent.

Vidu Q3 Image to Video is the still-to-motion row for this job: image required, native audio, 5s / 10s / 15s, 1080p, 16 credits, discounted. Transform the static piano into a short clip with diegetic keys. Prompt camera and physics only. Fingers that were already implied by an empty keyboard will be invented; keep hands out of the brief if you do not have a consented player in the still.

Do not describe "a crooner in a tuxedo singing our Friday standard." That sentence is a likeness and a composition you do not own. Do not describe the Yamaha as "a piano like ours." Re-describing the instrument is how the badge becomes a rumour.

Keep the piano large in frame. Fine badge type is the first thing a generator loses. If the house instrument must read as this grand, it has to fill the frame at the start. Fifteen seconds of the wrong lid is a long miss. Iterate at 5s.

Diegetic keys, not a vocal generate

Vidu Q3 has native audio. That is useful for key noise, room tone, a lid that actually sits in this bay. It is not useful as a singer. Write the mix you want: quiet keys, no vocal, no announced set list. A silent brief still gets a soundtrack you did not choose. Quoted lyrics in the prompt are how you get a mouth and a song you did not license.

Diegetic means the sound belongs to the object in frame. Keys on a locked piano are that object. A crooner is a second subject. Split the file if a real named singer is the night's offer: I2V for the instrument, a real take for the person, cut. Asking one generate to be the house Yamaha and a vocalist is how both fail.

Do not first-last the empty bench to a lifestyle still of a generated performer. The middle is a face. Likeness is not a piano-bar SKU.

Lifestyle around the room is a second clip. Cocktail bars already own the build. Do not restage that format on the grand.

A talking singer is a likeness you did not clear

A constructed vocalist is a private person the model invented. You cannot stand behind that face, that timbre, or that standard. If Friday needs a singer, put the real named singer on camera, with a release. Generation can still do the instrument. It cannot be the crooner.

Vidu Q3 from the still is the instrument job. Save talking models for a week when the mouth is a person who exists.

Overlay hours, cover, or the night's name with a text overlay after picture lock. Do not ask Vidu to typeset "FRIDAY 9PM" on the fallboard. Crop to 9:16 before you overlay.

Muted autoplay still needs the time on screen. Overlay does that. Auto-captions do not, unless someone spoke. A key bed is not a transcript.

Meter the room mix before you post

Vidu will give you audio with the picture. That mix is not a delivery stamp. The Loudness Meter never spends a credit. It is a browser file job: peak dBFS, whole-file RMS, and a BS.1770-style integrated estimate, labelled an estimate on purpose — not a certified EBU R128 meter. Drop the export. Read peak versus integrated. A file can be peak-limited and quiet; that pairing is the point.

Do not pay a video row to "make it podcast-loud" or "hit club level." The model will invent a mix. It will not give you a meter. Measure the file you already have. If the estimate is hot, fix it in an editor. If the keys drowned the room, reroll with a quieter brief, or score later and write silence in the generate.

Nothing uploads on the meter. The file stays on your machine. That is why it is free, and why it belongs after every Vidu take you might post next to a real room recording. A generated key bed that clips against the house PA is still a bad post.

Pause on the badge. Is it the still you would send as the room? Yes → keep the keys, meter the mix, overlay the hours. No → you prompted a singer or first-last'd a face onto the bench. Fix the still.

FAQ

Can I text-to-video "jazz singer at our Yamaha" and skip the room photo?

You can. You will get a bar and a face. It will not be this grand, and the face is a likeness. Vidu Q3 Image to Video exists because the still is the spec. Image required.

Does Vidu Q3 include native audio?

Yes. Audio is on: 5 / 10 / 15 seconds, 1080p, 16 credits, discounted. Write keys and no vocal, or write silence. A crooner line in the prompt is how you get a mouth.

Should I generate the Friday singer so the clip "has a hook"?

No. A generated singer is a likeness. Book the real named performer, or sell the room and the instrument. Overlay the set time.

Is the loudness meter a substitute for a house engineer?

No. It is a sanity check in the tab: peak, RMS, an integrated estimate. Use a real meter for a delivery spec. Use this page so you do not post a Vidu mix you never measured.