Two people, two photos, one stage
Since mid-September 2026, TikTok and X have been full of the same orange room: two people crowded around a single microphone that hangs from the ceiling, trading lines. Cartoon duos, footballers, couples, even cats. The scene comes from a real performance. In June 2022, COLORS released Quavo and Takeoff of Migos performing "Hotel Lobby", and that clip is the reference every AI version follows, which is why some people search for it as "Migos AI".
The hotel lobby AI video trend works differently from one-person recasts. You supply one photo per performer, and each photo is usually taken on a different day, with a different phone, under different light. The AI then has to place both people in one shot. When the two inputs disagree, the finished video shows it.
This guide is about closing that gap before you generate.
Where duo videos usually go wrong
Most disappointing results trace back to a mismatch between the two photos, not to a single bad picture:
- One person is a face crop, the other is half-body. The AI copies the outfit it can see, so one performer arrives fully dressed and the other gets clothing guessed from a collar.
- Different light. A warm kitchen selfie next to a cool office photo makes the pair look stitched together, even on the same orange set.
- A filter on one photo only. Beauty filters smooth skin and reshape features. Put a filtered face next to an unfiltered one and the difference is obvious.
- Wrong sides. The first photo you upload becomes the left performer and the second becomes the right. People often notice the order only after the video is finished.
- Expecting a height difference. Two separate photos do not tell the AI how tall each person is next to the other, so a real height gap will not show up.
The movement, camera and timing come from the reference performance in every case. Your photos control who appears, not what they do.
Picking a pair that already matches
Start by laying the two candidates side by side on your screen. You are looking for photos that could plausibly have been taken in the same session:
| Compare | Aim for |
|---|---|
| Framing | Both show the head, shoulders and upper body at a similar size |
| Direction | Both faces toward the camera, or turned only slightly |
| Light | Similar brightness and color temperature, no hard shadows |
| Clarity | Both sharp, unfiltered, with nothing covering the face |
| People | Exactly one adult in each photo, nobody else in the background |
Each file also has to be JPG, PNG or WebP, no more than 10 MB, and at least 512 pixels on each side. If you want the full list in one place, the two-photo hotel lobby maker's own photo checklist covers each point, including what two photos cannot show.
Framing is the one thing an edit cannot honestly fix. If one photo stops at the chin, retake it or choose another; asking an AI to invent the missing torso means inventing an outfit that person never wore.
FLUX Kontext edits that bring two photos together
The FLUX Kontext editor is useful here because it changes an existing photo from a short instruction instead of generating a new one. Upload a JPG, PNG or WebP of up to 5 MB, write one change per edit, and choose 3:4 as the aspect ratio. That gives an 896 × 1152 portrait, comfortably above the 512-pixel minimum and tall enough to keep the upper body. Each edit costs 10 credits.
Edit only the photo that is out of line, and keep the other as your reference. In every prompt, spell out that the person must stay the same.
Even out the color temperature
When one photo is orange from indoor bulbs and the other is blue from a window:
Neutralize the warm yellow color cast so the light looks like plain daylight. Do not change who this is: same face, hair, skin tone, expression and clothes.
Soften harsh light on one face
If one person was photographed under a ceiling spotlight or in direct sun:
Soften the hard shadows under the eyes and nose so the face is lit evenly from the front. The person's identity, features, skin tone and outfit must not change.
Remove colored party lighting
Pink and purple club lights read as skin color to a video model:
Remove the pink and purple party lighting and show the scene under neutral white light. Keep this person's face, hair, skin tone and clothing as they are.
Clear something in front of the outfit
A drink, bag strap or phone held across the chest hides the clothes you want carried into the video:
Remove the glass held in front of the chest and show the shirt underneath, continuing its color and pattern. Leave the face, hair and the rest of the outfit untouched.
After every edit, put the result next to the original. If the face has shifted at all, discard it. A slightly mismatched but genuine photo is better than a matched photo of someone who is no longer quite that person.
Edits to skip
- Do not change skin tone to match the other person. Each performer should look like themselves.
- Do not paint in eyes behind sunglasses or a face behind a hand. Retake the photo.
- Do not turn either person into a celebrity, including Quavo or Takeoff. The whole point of the trend is seeing someone new at that mic.
- Do not use a photo of someone who has not agreed. Both people should be adults who said yes to the video itself, not just to sharing a picture.
What it costs
Two parts of this workflow use credits, on two different sites:
- Kontext edits: 10 credits per edited photo on this site. If both photos already match, you spend nothing here.
- The Hotel Lobby video: on Hotel Lobby AI, uploading photos, previewing the left and right placement, and comparing options are free, but generating always uses credits. A 5-, 10- or 15-second video uses 30, 50 or 70 credits. The smallest pack is $9.99 for 100 credits, roughly $5 for one 10-second video, and there is no free trial or subscription.
Since a finished video that looks off still counts as a completed video, a couple of 10-credit edits up front are cheaper than a second take. The per-length numbers for every pack are listed on its pricing page; check it out before you buy, as prices can change.
What to expect from the finished clip
The download is a 720p landscape MP4, while TikTok and Reels favor vertical video. You can post it as is with bars above and below, or reframe it in an editor; if you crop, check that both faces stay in frame.
Choose Original audio if you want the familiar track. It is the template's vocals and music, not your voices, so lip sync can drift. Choose Silent video if you plan to add your own licensed sound. Both cost the same.
Before you post, watch the full clip and check that:
- each person is on the side you chose,
- neither face drifts back toward the original performers,
- outfits match the photos you uploaded.
Then label it as AI-generated. TikTok, YouTube and Instagram each ask creators to disclose realistic synthetic media, and a recast lands better when the audience is in on it.
The short version
Match the pair first, edit only what is out of line, never touch identity, and check the sides before you generate. The orange room does the rest.
Hotel Lobby AI has no connection to Quavo, the estate of Takeoff, Migos, COLORS or the song's rights holders, and none of them endorse it. Its prices, limits and steps above reflect aihotellobby.win as of October 16, 2026; they may change.