How We Built Hotel Lobby AI, Frame by Frame

This is the story of how we built Hotel Lobby AI, the generator behind this site. You upload two photos, and the Hotel Lobby AI video generator puts you and a friend (or your cat) into the orange booth to perform the routine everyone recognizes from the Hotel Lobby AI trend.

It took one weekend, five test rounds and 1,255 generation credits. Our first version looked fine and was wrong. Here is what we got wrong, how we fixed it, and why the result looks like the meme instead of a generic "two people rapping" clip.

A finished Hotel Lobby AI video from our generator. Both "performers" are AI-generated demo subjects.

Why we built a Hotel Lobby AI video generatorCopy Anchor Link

The Hotel Lobby trend started on September 16, 2026, when a TikTok edit swapped Quavo and Takeoff for Harvey and Mike from Suits. It passed 1.7 million views in twelve days. Two days later a version with two cats appeared on X; when we checked it, it had 7.8 million views. Search interest for "hotel lobby ai" peaked on September 26.

Plenty of tools already offered a Hotel Lobby AI video. Most of them used the same idea: put two people in an orange room and ask the model for a rap performance. We had already built a two-person photo-to-video pipeline for an earlier template, so we knew we could do the mechanics. The question was whether we could make a Hotel Lobby AI video that people would recognize in half a second.

How we built Hotel Lobby AI wrong the first timeCopy Anchor Link

Our first test copied the grammar most tools publish: a locked camera, a full-body shot, person A raps while person B nods, then they swap. The result was clean. Both faces held. Technically it passed.

Then we put it next to the real Hotel Lobby AI trend videos and it was obviously not the meme. The viral clips are not locked off and nobody stands still. Our "safe" prompt had produced a polite karaoke clip.

So we stopped reading prompts and started watching footage.

Watching the Hotel Lobby trend frame by frameCopy Anchor Link

We stepped through the original COLORS show and three of the biggest remixes — Suits, Taylor Swift and Travis Kelce, and the two cats — one second at a time. Three different creators, on different platforms, with humans and cats, and the movements land on the same second in every one. The Hotel Lobby trend is really a fixed 15-second routine:

That table became our Hotel Lobby AI video template. It also explains why the trend works with the official sound: when everyone uses the same routine and the same 15-second clip of audio, the moves fall on the same words every time. Ten seconds would cut off the hush, which is the payoff.

We also noticed details our first version missed: the microphone is silver, not black; the orange wall glows in the center and darkens at the edges; and the two performers stand close, turned toward each other, sharing one mic at head height.

Picking the model on cost, not hypeCopy Anchor Link

The tool that popularized the trend recommends ByteDance's Seedance 2.5. We priced it before using it. Through our API provider, Seedance 2.5 costs 63 credits per second at 720p — a 15-second Hotel Lobby AI video would cost more than we charge for it. Seedance 2.0 Mini costs 8.2 credits per second, so a 15-second clip is 123 credits.

We had already used 2.0 Mini in reference mode for our previous template, and it kept two faces apart there. We chose it, and spent the money on getting the routine right instead.

For the free preview image we use GPT Image 2.5: it composes your two photos into the booth in about a minute.

The bugs we fixedCopy Anchor Link

Rewriting the prompt as four timed shots worked on the first try — the point, the explanation, the spin and the hush all appeared within about half a second of the template. Then the details started to show:

• A gray background leaked in. For about 1.5 seconds of one wide shot, the orange booth turned gray. The model had borrowed the plain backdrop from the reference photos. One sentence telling it to take only the people from the photos fixed it, and later held up with café and living-room selfies.
• Shots that were too wide. Some shots showed tiny full-body figures. The viral clips are knees-up. We told it never to go full-body wide.
• A white preview. The free preview image kept coming back on a white wall. Describing the orange wall in words, and dropping one API parameter, fixed it.
• Cats on all fours. In the cat version, the cats sat, and the right one dropped to all fours to turn around. Funny, but not the joke. The joke is cats performing like people. "Stay upright on the hind legs for the whole video, even while turning" fixed it for cats and dogs — you can see the result in our Hotel Lobby AI cats guide.

We also had to decide what not to check automatically. We only reject technical failures: a failed render, a broken file, black frames or frozen frames. We don't score faces, because face scores raised false alarms in our earlier project. Instead, every template change gets watched frame by frame by a person.

Vertical or horizontal?Copy Anchor Link

Every viral Hotel Lobby AI video we measured was landscape — the format comes from a YouTube show. We assumed vertical would squeeze two people facing each other. It didn't: in 9:16 the pair fills the frame and the close-ups get bigger faces. So you choose. Vertical is the default for TikTok and Reels; horizontal looks like the original show.

The same routine in landscape.

How we built Hotel Lobby AI, by the numbersCopy Anchor Link

• 5 test rounds plus a batch of showcase videos, 1,255 credits in total — about $12.55 at our most expensive top-up rate.
• Generation time: 161 to 267 seconds per 15-second video; the preview image usually takes one to two minutes.
• Tested and working: two very different people, half-body phone photos with busy backgrounds and glasses, two cats, a dog and a cat, and a person with a pet.
• Not tested yet: two people who look alike (the hardest case for keeping faces apart), and very side-on photos. We say so on the generator page.

We built the pipeline with help from an AI coding assistant (Claude Code). The test photos and showcase subjects are all AI-generated and fictional.

What this means when you use itCopy Anchor Link

You don't write a prompt. You upload two photos, choose person or pet for each side, pick vertical or horizontal, and get a free watermarked preview. If you like it, one 15-second Hotel Lobby AI video costs $4.99. If a render fails for technical reasons, your credit comes back.

Then add the official sound when you post — our guide on how to make a Hotel Lobby AI video shows how to line it up, and what the Hotel Lobby AI trend is covers where it came from.

Make your Hotel Lobby AI video →