About this trend

This is the prompt our generator sends: "Two performers stand side by side in a seamless, evenly lit, saturated orange studio with an orange floor and no props. A single black microphone hangs from the ceiling on a cable between them at head height. The performer on the LEFT is the person from @Image 1. The performer on the RIGHT is the person from @Image 2. Keep each face, gender, age, hairstyle, skin tone, body type and outfit exactly as in its reference photo for the entire video; glasses, hats and facial hair only where the photo shows them. The two faces never merge or swap sides, no third person appears, and anything the people hold in their photos is left out. They take turns rapping into the microphone: while one leans in and raps with confident hand gestures, the other nods, bounces to the beat and reacts, then they swap. Static front-facing camera, full-body wide shot, both performers fully in frame, no cuts, no zoom, no slow motion, no text, no watermark. Energetic original hip-hop beat." The generator adds one more line, a short description of each person's hair, glasses and outfit read from the uploaded photos. Why it is written this way: the scene is described first and kept bare (no props, one light colour) so the model does not invent extra objects; each reference photo is tied to one side so the two people do not swap places; the identity line asks for the whole video because faces tend to drift over a long clip, and the next line rules out the usual failures (merged faces, a third person, a phone from the photo showing up); and the camera line forbids cuts and zooms, which is what keeps the clip looking like the original single-take performance.

FAQ

Which AI model does this prompt work with?

It is written for Seedance 2, which takes up to nine reference photos addressed as @Image 1, @Image 2. Other models that accept two reference images work too; rename the references to their syntax.

Why does my result have the wrong faces?

Use one person per photo, front-facing and well lit, and keep the reference order the same as in the prompt: left is image 1, right is image 2.

Can I add the original song with the prompt?

No model should be given the commercial recording. Generate with an original beat and add the trending sound inside TikTok, Reels or Shorts.

Is it cheaper to run the prompt myself?

It can be if you already pay for a model with reference images. On this page you pay per finished video and failed runs are refunded.