Skip to main content
All posts
Guide

How to Make the 58th Floor AI Video From Two Photos

A step-by-step guide to the 58th Floor AI video: who goes on the left and right, which photos work, pets, posting with the sound, and how to fix a face that comes out wrong.

The 58th Floor AI video is the trend where two people trade the hook and the verse of Anderson .Paak and Cordae's "58th Floor" at one microphone hanging in a teal studio. You do not need to film anything, own a teal backdrop or unlock a CapCut template. You need two photos.

This guide walks through the whole thing: picking the photos, deciding who raps what, making the clip, posting it, and fixing the usual problems.

What you need

  • Two photos, one person per photo. Separate selfies taken in different cities are fine.
  • Permission from everyone pictured. Everyone must be 18 or older, and no celebrities or public figures.
  • A phone or computer browser. The studio on the 58th Floor AI homepage runs in the browser; there is no app to install.

If you only have one photo of the two of you together, that works too: the studio lets you tap each face and cuts each person out as their own portrait.

Step 1: Decide who takes the hook and who takes the verse

The clip has two places at the mic and they are not equal.

PlaceWhat they doScreen time
LeftRaps the end of the hook to camera in the opening two-shot, then listens and nodsAbout 5.5 seconds as the lead, then a supporting role
RightTurns in, takes the verse, gets the close-up, the wide and the face-offAbout 9.5 seconds as the lead

So the rule is simple: put the bigger personality, or the person the video is about, on the right. If it is a birthday post, the birthday person goes on the right.

Step 2: Pick photos the model can read

The AI copies what it sees. A good photo gets you a good video; a bad photo gets you a stranger.

Photos that work:

  • One face per photo, sharp, looking roughly at the camera
  • Framed from the waist up, so the outfit comes along into the studio
  • Daylight or a bright room, no beauty filter
  • Hair and face not covered

Photos that cause trouble:

  • Sunglasses, masks, a hand over the face
  • Group shots where the face is tiny
  • Heavy filters or strong colour casts
  • Very old, very blurry or very dark pictures

The right-hand photo matters most, because it gets the close-up. Use your sharpest picture there.

Step 3: Make the clip

  1. Open the studio and choose Duo (or Pets, see below).
  2. Tap the left side of the mic and add the hook photo, then the right side for the verse photo.
  3. Choose Standard (720p) or HD (1080p).
  4. Tick the box confirming you may use the photos, and press the button.

The video is ready in about three to eight minutes. You get a 15-second, 9:16 MP4 with no watermark. It keeps the live recording from the COLORS session, so every mouth move and cut lands on the beat.

Pets on the 58th Floor

Choose Pets and the two performers become your animals. The pet on the left wears the flat cap and T-shirt and finishes the hook; the pet on the right wears the black jacket and takes the verse. Dogs and cats both work. As with people, use one animal per photo with the face clearly in view.

Step 4: Post it

  • TikTok: upload the MP4, and turn on the AI-generated content label under More options before posting. TikTok asks for it on realistic AI video.
  • Instagram Reels and YouTube Shorts: upload as a vertical video; it is already 9:16.
  • If the app mutes the audio: the clip carries the song, but some apps strip or block music on upload. Add the trend sound, "original sound - COLORS", from any 58th Floor video and line it up with the opening of your clip.

The lyrics are explicit, so think about where you post it.

Fixes for common problems

A face does not look like the person. It is almost always the photo. Try a sharper, front-facing picture in good light, and avoid photos where the person is small in the frame.

The wrong person got the close-up. The close-up always goes to the right-hand photo. Swap the two photos and make it again.

An outfit changed. The model takes the outfit from the photo. If only a face is visible, it has to guess the rest; a waist-up photo fixes this.

Someone is wearing a cap or sunglasses they do not own. The studio tells the model not to add the original performers' cap, chain or sunglasses, but a blurry photo can still confuse it. A clearer photo helps.

The render failed. When a generation is confirmed as failed, its credits go back to your balance automatically.

What it costs

A Standard video is 10 credits and an HD video is 20. Credit packs are a one-time payment: $9.80 for one Standard video, with bigger packs on the pricing page. Credits never expire and there is no subscription.

Keep reading

Make your own 58th Floor AI video

Add two photos and get the 15-second vertical video with the original sound in about three to eight minutes. From $9.80, no subscription.

Add your two photos