How to Make an AI Rap Video From Two Photos

Three inputs, and the model draws everything else — the second performer, the room, the beat.

• Two portraits — one for the left performer, one for the right. Front-facing, well lit, one person per photo, and photos you have the right to use.
• A scene — two presets are built in (a night alley and a studio booth), or upload a photo of your own place.
• Four bars — your lyrics. A set is prefilled so you can render first and write later.

File limits, so uploads do not bounce: JPG, PNG or WebP, up to 30 MB each, and every edge of the image must land between 256 and 5760 pixels — a normal phone photo clears both with room to spare.

The left slot becomes the performer on the left of the frame; the right slot, the one on the right. That is the whole trick and the easiest thing to get wrong: if the duet feels swapped — voices and faces not where you pictured them — swap the two photos before you blame the model.

What makes a good portrait here is the same as what makes a good passport photo: face toward the camera, even light, nothing else in the shot competing for attention. Sunglasses, heavy shadows across the face, or a group photo cropped down all give the model less to hold on to, and it shows in the lip-sync.

Two presets ship with the studio: a night alley and a studio booth. Pick one and you are done — the set, the lighting and the mood come with it.

Or drop in your own background photo. The model rebuilds the scene around it, which is the difference between a clip that reads as your city and one that reads as a stock stage. Busy backgrounds work, but give the eye one clear anchor — a doorway, a sign, a skyline — rather than a crowd.

Lyrics get their own box, separate from the director prompt, for a practical reason: you can rewrite one without touching the other. A set of four bars is prefilled so the very first render works with zero writing, and a swap button brings in another set — four are stocked, each with its own mood — without disturbing anything else you have changed.

When you write your own, four short lines are the format. Each line lands on a beat, the two performers trade them, and short lines survive the lip-sync better than dense ones. The full prompt — direction, scene, and your lyrics appended at the end — can run to 7,000 characters, which is far more than a duet needs; the limit is there so ambitious directors are not silently truncated.

Pressing generate starts a cloud render — usually a few minutes, and you can leave the tab and come back. You will need an account first; if you are not logged in, a sign-in window opens before anything is sent.

When the clip lands it plays right on the page with a download button under it. If a render fails, the reason comes back in words, not a code — photo too small, prompt too long — and fixing that one thing is usually all the retry needs.

You get a vertical MP4, finished — vocals, lip-sync, scene and all. It carries an AI-generated label and an identifier inside the file, which is what TikTok, Instagram and YouTube policies and the EU AI Act ask for. Keep the label when you post; cropping it out is the part that breaks those rules.

Results vary run to run — the lip-sync most of all. A second render with the same three photos often lands better than the first, and since a failed run is free, the cheap experiment is one more try, not a new plan.

• Portraits should be people whose photos you may use — your own, or with permission. It is a hard line, not a suggestion.
• Lyrics you write are yours to clear: a borrowed line from a copyrighted song is a copyright question the render does not answer for you.
• RapReel is an independent, fan-made tool and is not affiliated with the model vendors behind it.