58th Floor · COLORS duet

58th Floor AI Video: Rap the COLORS Duet With a Friend

Upload one photo of each person and the AI puts both faces on the two performers in the teal studio duet — one on the verse, one on the hook, mouths synced to the vocal, in the wide framing the clip is known for. Credits to start · No filming · Crop or letterbox for Reels, Shorts and TikTok after export.

The two-person studio preset is loaded by default. Remove it to use your own 3–15s motion reference.

Reference performance

The Hotel Lobby preset is loaded by default. Remove it to use your own 3–15s motion reference.

Preset

First person

Required · left performer, or a photo containing both subjects.

Second person

Optional · right performer when using separate photos.

—

The Clip You Get

A short clip of two people trading bars in a teal-lit room with a single microphone hanging between them: both faces replaced, both mouths moving on the part each performer is assigned, cut in the wide framing of the source performance.

What changes is who is standing in the room. The room, the light, the mic and the blocking stay put — that is the whole appeal of the format: viewers recognise the staging in half a second, then notice it is their friends inside it.

Faces are swapped, not generated from scratch, which is why the output keeps the head movement and the shoulder work of the original take. A still photo dropped on a static body reads as a filter; this reads as performance.

What the 58th Floor Trend Actually Is

For anyone who landed here from the song: "58th Floor" is the Anderson .Paak and Cordae track performed as a duet on COLORS — a small teal-green room, one hanging microphone shared between the two of them, a wide shot with no cuts to hide behind. The clip works because the two performances are genuinely different: one carries the verse, the other answers on the hook, and both are moving the whole time.

That is exactly the shape a face swap needs. The trend then did what these trends do — people wanted themselves and a friend in that room. The version you are about to make is a recreation of the staging, not a copy of the footage: two photos in, one duet out.

Two Photos, Two Performers

1. Pick two photos that can share one room

Both faces shot in similar light, from a similar angle, cropped hairline to chin. A daylight selfie next to a photo taken in a dark bar is the single most common reason a duo swap looks fake: the model corrects exposure per face, and the two corrections do not match.

2. Decide who takes the verse and who takes the hook

The uploader asks which photo goes on the left performer and which on the right, and that assignment decides who mouths the opening bar. Get it backwards and the whole verse lands on the wrong person. Two friends: give the verse to whoever talks more, because the hook reads as a reaction. A couple: the answering part usually has the better line, so put it on whoever is likelier to have the last word — which is why so many of these are funny.

3. Generate, then frame it for where you post

Output follows the studio staging's wide framing. Regenerating will not change the native aspect ratio, so the work of going tall happens after export — read the framing section below before you post.

Why This Scene Reads as Rap

The room is built as a performance, not a photo set: one mic hanging between two people instead of two separate marks, a stage-wide framing, and a single coloured light source giving both faces the same direction. That combination is what makes a replaced face read as performing rather than pasted in. It also explains why the format travels with rap audio specifically — both performers in the source are already delivering bars, heads moving, mouths working, which is the exact motion a swap needs to stay alive.

The Wide-Shot Problem: Getting It Into 9:16

COLORS is a wide composition: the two performers sit toward the left and right thirds with the microphone hanging between them — the worst possible layout for a straight vertical crop. Pull a 9:16 window out of the centre and you keep the mic and cut both people in half. Three ways to handle it, best-looking first:

  1. Crop with both performers in mind. Pull a tall window that still includes both heads and the mic, even if you lose side wall. Better than a dead-centre crop that halves both faces.
  2. Pan between the two performers. Keep the wide frame and cut a moving tall window that follows whoever is rapping, switching on the bar. It hides the crop, but needs a cut per line and looks busy if the switch is off by a beat.
  3. Letterbox the wide shot. Blur a copy of the frame behind a centred wide video and stack it into a tall canvas. Fast and safe, and it clearly reads as a recycled wide clip.

Word-by-word captions sit in the empty space above the performers' heads — the one area a tall crop can safely cover. That is why a careful crop is an advantage here rather than a compromise.

The 15-Second Cut and the 30-Second Extended Session

A single render follows the length of the reference you give it (3–15 seconds). The standard version of this trend is one 15-second pass: verse and hook in one take, ending on the hook so the last frame flows back into the first and viewers watch twice without noticing.

The 30-second extended cut going around is two takes, not one: render the verse from your two photos, render the hook from the same two photos, then cut them together on the beat. Keep the photo assignment identical across both renders — the lip sync is baked into each one, so a swapped assignment lands the wrong mouth on the join.

The Audio Question

Every version of this trend you have seen runs on the same audio, and that audio is a commercial recording. That matters more than it looks:

  • Posting it on a social platform generally falls under that platform's music licensing for organic posts, and the platform may mute, limit or share revenue on the post. That is not a licence for anything else.
  • Using it in an ad, a brand video or anything you are paid for needs clearance you probably do not have.
  • Re-uploading the clip with the original audio to a site you monetise is the riskiest use of the three.

The clean route is to generate or upload your own track and swap it in. The staging does not care what is playing — the mouth movement is generated against whatever audio you supply — so a track you own gives you the same format with none of the exposure. If you want the original feel, ask for the same tempo and delivery, not the same recording.

Two Faces That Look Like They Belong Together

  • Same light — both daylight, or both indoors under the same kind of lamp.
  • Same angle — both facing the lens; one profile shot breaks the pairing.
  • Same crop — hairline to chin for both.
  • Similar sharpness — one soft filtered photo next to a crisp one reads as two videos stitched together.
  • No sunglasses, no hats pulled low, no heavy beauty filters.

If the result looks off, replace the weaker photo before you regenerate. A better photo fixes more than any setting will.

Common Mistakes

  • Uploading the weaker photo and hoping the model compensates — it does not; reshoot in daylight.
  • Assigning the verse to the quieter person, so the opening bar sits on the wrong face.
  • Posting the wide export straight to Reels, where both performers end up as slivers at the edges.
  • Using the original track in a paid ad.
  • Rendering the extended cut in one pass and losing the beat at the join.

FAQ

What is the 58th Floor AI video?

A photo-to-video recreation of the teal-studio duet staging from Anderson .Paak and Cordae's COLORS performance. Two photos go in, two faces come out, lips synced to the bar.

How do I make a 58th Floor AI video with two people?

Upload one photo of each person, assign each face to the left or right performer, and generate. The AI replaces both faces and syncs each mouth to the part that performer delivers.

Do I need the original COLORS clip?

No. You are recreating the staging — teal room, hanging mic, wide two-shot — not re-uploading the footage. Use the two-person studio preset, or upload your own 3–15s reference if you have footage you have the rights to.

Can I make it vertical for TikTok, Reels and Shorts?

The scene exports wide to match the staging. For tall feeds, crop keeping both performers in frame, pan between them on the bar, or letterbox the wide shot; a centre crop usually cuts a performer off.

Is the 58th Floor AI video free?

Generation uses credits. You can load the preset and choose photos before spending any, and exports carry no watermark.

Can I use the original song in the video?

For organic posts a platform's music licence usually covers you; for ads, brand work or anything you are paid for, swap in a track you own.

Can I change who raps first?

Swap the two uploads and regenerate. The lip sync is baked into the render, so the assignment has to be right before you press generate.

Why does my version look fake?

Almost always the photo pair: mismatched lighting, a turned head or different crops. Replace the weaker photo first.

Make Your 58th Floor Duet

Two photos, one studio, fifteen seconds. Upload both faces, pick who takes the verse, and generate.

Generate the 58th Floor duo clip