Make another take of your AI video when the result is close and the problem is small and random, such as one odd hand or a face that drifts for a moment. If the same problem shows up again, or the result is clearly off from the start, change something first: the photo's card, "Keep the original look", the photo itself, the part of the video or the quality. Every take is a new render with its own price, so spend it on a setup that has a real chance to do better.
What exactly is "another take"?
On vswp, pressing "Create video" again with the same video and photos makes a new take. The AI generates the whole part again from scratch, so two takes never come out exactly the same: the face, the hands and small details in motion are drawn anew each time.
That's why a take can fix a random slip. It's also why it can't fix a problem built into the setup: if the photo is soft or sits on the wrong card, the next take starts from the same weak point.
When is a plain new take worth it?
A new take with nothing changed makes sense when all three are true:
- The result is mostly right: the person reads as you or your friend, the outfit and hair are what you expected.
- The problem is short and local: a hand that looks wrong for a second, a brief blink of the wrong expression, a face that slips during one fast turn.
- You've already checked the photo, the cards and the settings.
In this case the cause is often plain chance, and a fresh generation can land differently.
When should you change something first?
Change something before the next take when:
- the problem is there for most of the video, not one moment;
- two takes in a row miss in the same way;
- the wrong person was replaced, or someone you gave a photo to stayed as in the original;
- the clothes or hair are not what you meant.
Repeating the same setup in these cases usually repeats the same result.
What to change for each problem
| What you see | What to change first |
|---|---|
| The face doesn't look like the person | The photo: sharper, unfiltered, larger face, angle close to the video |
| Your face is on the wrong person | The photo's card |
| Two people look the same | Give each person in the frame their own photo |
| Wrong clothes or hair | "Keep the original look": on keeps the video's clothes, off takes the look from a full-body photo |
| Face slips in fast moves or a crowd | A calmer part of the video |
| Face too soft because it's small | 720p instead of 480p, or a part where the person is closer |
| Someone wasn't replaced | Check they are on a card; people seen briefly or far away are grouped as the background |
Our guide on what to do when a face swap doesn't look like you goes deeper into the photo fixes, and who gets swapped in a video explains the background group.
In what order should you check things?
Start with what costs nothing to check and is most often the cause:
- The cards. Is each photo on the person you meant? In a video with several people, this is the first thing to look at.
- "Keep the original look". On, only faces and hair change and the clothes stay as in the video. Off, the clothes, hair and silhouette come from the photo, which works best with a full-body photo of one person. A portrait can't show clothes it doesn't contain.
- The photo. One person, sharp eyes, even light, no filters, the original file rather than a forwarded copy. See the best photos for an AI video swap.
- The part of the video. A new video covers up to 30 seconds. In a longer clip, pick a part where the person is closer and moves more calmly; picking the best 30 seconds shows how.
- The quality. If the face is small throughout, 720p keeps more detail.
- The video itself. If the face is covered, tiny or the camera never stops shaking, a different video may serve you better than any number of takes.
Change one or two things at a time. Then, if the next take is better, you know what helped.
How do you keep takes from adding up?
Each take is a new render. The price is in Telegram Stars (⭐), depends on the length, the quality and the number of people, and the app shows it before you start. A few habits help:
- Test on a shorter part. If you're not sure the photo works, try it on a short stretch where the problem showed up before you redo the full part.
- Use 480p while testing, then make the final take at 720p if you need the detail.
- Use your free preview first if you're new: about 5 seconds of your video at 480p with a small "vswp.app" mark, from where the people with photos are on screen.
- Know what's covered. Stars come back automatically when a render fails or is canceled. A finished video is a completed render, even if you want another take; see what happens to your Stars.
When is it time to stop?
Generative AI is not frame-exact editing. Hands, things in front of the face, very fast camera motion and crowded scenes can stay tricky even after a few takes. If two or three takes with a better setup still miss in the same place, the clip itself is the limit. Pick a different part, or a calmer video, ideally one from Trends, where the people are already found. More on that in hands, fast moves and crowds in an AI video swap.
Whatever you change, use only videos and photos you have the right to use, and ask before you use someone else's photo.
FAQ
Does another take with the same photo give a different result?
Yes. Every take is generated anew, so the face, hands and small details come out a little different each time, even with the same video and photo.
Do I pay for another take?
Yes, each take is a new render with its own price in Stars, shown before you start. Stars come back automatically only if a render fails or is canceled.
How many takes should I try before changing something?
If a take is close and the problem is a single moment, one more take is reasonable. If two takes miss in the same way, change the photo, the card, the setting or the part before trying again.
Can I fix just one second of the video?
A new take covers the whole part you chose. If the trouble is in one short stretch, choosing a part that leaves out that moment is often the simpler fix.