How to Make a Migos AI Video from Two Photos

To make the two-person rap clip people call a Migos AI or Hotel Lobby AI video, start with two separate portraits, assign one person to the left and one to the right, and ask for one short call-and-response performance. The best first attempt is ten seconds, a fixed camera, a single microphone, and no more than one gesture per person.

That is enough structure to feel like a performance without asking a short generation to solve a music video’s worth of movement.

Our Migos AI Video Generator makes a new H3 Max scene from the two portraits and your direction. It does not paste faces into the official “Hotel Lobby” footage. The beat, lines, motion, and booth are generated for your new clip.

Prepare two photos that can hold their roles

Use one photo for each performer. A good reference shows the whole face, natural skin texture, visible hair, and roughly front-facing eyes. It does not need to be a studio portrait; it does need to make identity unambiguous.

Avoid these inputs on the first try:

• a group photo cropped twice;
• sunglasses, hands, or a phone covering the face;
• a profile view for one person and a front view for the other;
• strong beauty filters that erase facial detail;
• tiny faces inside a full-body photo.

Put the person who should begin the verse in Image 1. They will be described as the performer on the left. Put the answering performer in Image 2, on the right. This explicit mapping is more reliable than “the man in black” or “my friend,” because clothing can change during generation.

Pick one exchange, not a whole song

A useful ten-second plan has three beats:

1. both people settle into the rhythm;
2. Image 1 delivers one short original line and Image 2 answers;
3. they share one final look, nod, or fist bump.

If you add walking, a spinning camera, a wardrobe change, a dance break, and four lines of dialogue, the model has to trade identity stability for motion. Keep the camera fixed until the faces are working. You can make a more ambitious second clip later.

Here is a complete starting direction:

Image 1 performs on the left and Image 2 performs on the right.
Place them in an original warm-orange recording booth with one
hanging studio microphone between them. Fixed vertical medium-wide
camera, both people visible from the waist up. They nod to an original
minimal hip-hop beat. Image 1 delivers one short original line, Image 2
answers with a different short original line, then both face the camera
for the last beat. Preserve each person's face, hairstyle, skin tone and
clothing from their own photo; do not merge or swap identities. Natural
hand gestures only. Generate an original beat and original vocal phrases.
No existing song, artist voice, celebrity likeness, logos, captions or text.

You can load that setup directly with the orange-booth scene, then change the original lines or occasion before generating.

Choose ten seconds before fifteen

Five seconds works for a single shared reaction, but it is tight for two vocal turns. Ten seconds gives each person a moment without leaving a long idle ending. Fifteen seconds is useful only when the prompt has a clear third beat.

The generator offers 480P and 768P. Use 480P to test the role mapping and action at a lower credit cost. Move to 768P after the composition works. A larger file does not repair an unclear reference photo or an overloaded prompt.

The current reference-video rate is shown beside the button: 8 credits per second at 480P and 12 credits per second at 768P. If you do not already have enough credits, the page also shows a one-video price for the exact duration and resolution you selected. Payment adds those credits; it does not start the generation until you return, review the scene, and press Generate.

Review the result in the right order

Watch once with the sound muted. Check whether Image 1 stays on the left, Image 2 stays on the right, and each face remains recognizable through the final frame. Then listen without scrutinizing the picture. Is each vocal turn short enough to land? Does the beat leave room for the words?

If the faces swap, make the role sentence the first line of the prompt and remove unnecessary movement. If a hand distorts, replace pointing or finger-counting with a nod or open-palm gesture. If the voices overlap, use “Image 1 speaks first. After a brief pause, Image 2 answers.” The Hotel Lobby AI prompt guide shows how to tighten each of those instructions.

For a social post that uses a licensed platform track, export the visual, mute or lower the generated audio in your editor, and add music through the platform’s own authorized library. The song and trend guide explains why the music inside a reference clip and the music licensed for a post are separate decisions.

The trick is not to imitate every detail of a video you have seen. It is to keep the instantly readable setup—two people, one mic, one exchange—then give those two people a line that could only belong to them.