The best part of a long video for an AI swap is the stretch where the people you replace are large, well lit and facing the camera, the camera is calm, and the moment you care about starts and ends inside 30 seconds. Look for it before you render, not after: the first 30 seconds are often an intro, and a shorter, well-chosen part usually looks better and costs less than the first one that comes along.
Why can't the whole video be swapped at once?
On vswp you can upload a video from 1 to 120 seconds long, up to 150 MB, in any common format. Each new video is made from one part of up to 30 seconds, with the original sound of that part.
For an upload, the search for people runs automatically on the first 30 seconds, usually in a couple of minutes, and it's free. If the moment you want comes later, choose another part of up to 30 seconds. If your file is longer than two minutes, trim it to 120 seconds or less before uploading.
What makes one part better than another?
Inside one video, quality can change a lot from second to second. A dance clip may open with a wide shot from across the room and then cut to a close-up. The swap redraws the people in every frame, so the part you pick sets how much the AI can see.
Watch the clip once just for faces, and note the seconds where:
- The faces are large. The people you replace fill a good part of the frame, not small figures far away.
- The faces are in view. Mostly toward the camera or in a gentle turn, not long stretches from behind or in profile.
- The light is clear. No dark corners, strobes or heavy colored light on the faces.
- The camera is calm. Steady or smooth, without shaky handheld moves or whip pans.
- Nothing covers the face. Hands, hair, drinks and microphones pass by for a moment, not for the whole scene.
- The cast is small. One or a few main people, rather than a crowd where people overlap.
The more of these a part has, the fewer takes it is likely to need. Our guide on how to choose a video for an AI swap goes through each point in more detail.
How do you find the right moment?
A good part is also a good clip on its own. Someone who scrolls past should get it without the two minutes around it.
- Name the moment. Write down what the clip is for: the punchline, the big move, the chorus, the goal. Everything in the part should lead to it.
- Find its timestamp. Note where the moment ends, then count back. If it ends at 1:12, the part can start no earlier than 0:42.
- Cut the warm-up. Start close to the action. A few seconds of setup are enough; long intros, logos and people getting into position can go.
- Check the faces in that window. Use the list above. If the faces are tiny for the first ten seconds, start later.
- End on a beat. Stop on the last word of the joke, the end of a musical phrase, or the landing of the move.
Does the sound matter?
Yes. The result keeps the original sound of the part you choose, so your cut points are also audio cuts. A part that starts mid-word or ends halfway through a line feels broken even when the faces look great.
Should you always use the full 30 seconds?
No. Thirty seconds is the upper limit, not a target. Many trends and memes take well under that, and a shorter part has real advantages:
| Shorter part | Longer part (up to 30 s) |
|---|---|
| Costs less, since price depends on length | Room for a story with setup and payoff |
| Easier to keep only the strong seconds | More chances for a weak stretch to slip in |
| Fits short loops in stories and Reels | Fits a full chorus or a longer scene |
If the best faces cover 12 seconds and the next 18 are a wide shot of the crowd, a 12-second part is usually the better video.
What if the best moment is spread over the whole video?
Sometimes the joke is at the start and the best move is near the end. You can:
- Pick the one that matters most. One strong moment usually beats two half-moments.
- Make two videos. Each part is its own render with its own price; post them as a series.
- Shorten your own footage first. Cut the slow middle out before uploading, if it's a video you have the right to edit and use.
How do you test a part before paying?
Every new user gets one free preview: about 5 seconds of the video at 480p, taken from where the people with photos are on screen, with a small "vswp.app" mark. It shows how your photo works in your chosen part, and if it fails, the free try comes back. More in what's free on vswp.
The full video is paid in Telegram Stars (⭐). The price depends on the length, the quality (480p or 720p) and the number of people, and the app shows the exact price and the estimated time before you start.
What if the result still isn't right?
If faces look off in one stretch, that stretch is often the reason: a turn away from the camera, a fast move, a hand over the face. Try a part that skips it. Each press of "Create video" makes a new take, which is a new render; this is generative AI, not frame-exact editing. See what to do when a face swap doesn't look like you.
FAQ
Why are people only found in the first 30 seconds?
The automatic search covers the first 30 seconds of an upload, since one video is made from one part of up to 30 seconds. If your moment is later, choose that part instead.
Can I make a video longer than 30 seconds?
Not in one render. You can make several videos from different parts of the same clip and post them one after another.
Is a shorter part cheaper?
Yes. The price depends on the length, the quality and the number of people, so a shorter part costs less. The exact price shows before you start.