How Long a Face Swap Takes, and What Happens While You Wait
- Work is measured in frames, not megabytes. A photo is one frame, a GIF is tens, a video is hundreds or thousands.
- A photo swap usually lands inside a minute. A GIF takes a minute or two. A video runs for minutes, and a longer clip runs longer.
- Free jobs share one processing lane, so your place in the line gets added to your own processing time.
- Closing the tab does not stop the job. It keeps running on the server and appears in My Files when it finishes.
- A job can fail late because the swap server has a time budget of several minutes. Tokens go back when that happens.
You uploaded a clip four minutes ago. The bar has not moved since you looked. Nothing on screen says whether that is normal or broken. So you close the tab, and now you do not know if you killed the job or just the page.
Frames are the unit of work, not megabytes
No tool swaps a clip in one shot. It pulls the clip apart into single frames. On every frame it finds the face, lines up the angle, blends your face in, then cleans the edges. Only after that does it put the frames back together.
So the real question is simple. How many frames did you just hand it?
Multiply the length by the frame rate. Ten seconds at 30fps is 300 frames. The same ten seconds at 24fps is 240. That is 240 separate detect and blend passes for one short clip.
File size barely matters here. A tight 2 MB clip that runs 20 seconds is far more work than a fat 8 MB clip that runs 3 seconds. The Face Swap Video tool charges you time by frame count, not by download size.
Work out your own frame count before you upload
Do the maths first and the wait stops feeling random. Here is how the three formats compare, with the caps a free account gets.
| What you upload | Frames the model runs on | What sets the wait | Free account cap |
|---|---|---|---|
| Photo | 1 | One detect, align and blend pass | 2 MB image |
| GIF | Roughly 20 to 150 | Frame count, plus rebuilding the colour palette | 150 frames, 5 MB |
| Short video, 10 seconds | 240 to 300 | Length times frame rate | 20 seconds, 5 MB |
| Longer video, 60 seconds | 1,440 to 1,800 | Same maths, six times the frames | Pro only, up to 100 MB |
Anonymous users get tighter caps: 8 seconds of video and 80 GIF frames. Those caps exist because of the queue, not to annoy you. One long clip can hold the free lane for minutes.
Illustrative. These are the fallback per-format estimates the queue falls back on before it has enough recent jobs to average. The live estimate you see on screen comes from a rolling average of recent finished jobs, so your number will differ.
One free lane, and where you stand in it
Processing time is only half your wait. The other half is the queue.
Free jobs share a single processing slot. Your job sits as "queued" until the job ahead of it finishes, then it flips to active and starts burning frames. That is why two people can upload the same clip and wait very different amounts of time.
The status panel shows your position and an estimate. The estimate is your position multiplied by the recent average for that format. It uses real finished jobs from the last few days, so it moves with how busy things are.
Paid plans skip the free lane entirely and start straight away. If you upload clips often, that queue jump is the main thing you are buying on the Pro and Max plans, alongside longer clips and bigger files.
The GIF bar counts frames, the video bar estimates
These two bars mean different things, and mixing them up causes most of the panic.
A GIF bar is a real counter. It reports frames finished out of frames total. When it moves, actual work happened. When it sits still for a long stretch on the GIF Face Swap tool, something genuinely stalled.
A video bar is an estimate. It compares time elapsed against the average job of that kind. Your clip might be longer or busier than average, so the bar can run ahead of the truth. It stops at 95% on purpose and only jumps to 100% when the file is really ready.
A video sitting at 95% is not frozen. It just outran the average. Give it another minute or two.
Six things that stretch the wait
- A longer clip. Every extra second adds 24 to 60 more frames.
- A higher frame rate. 60fps doubles the work of 30fps at the same length.
- A higher resolution. Bigger frames take longer to detect and blend.
- More faces in shot. Each face gets its own detect and align pass.
- Extra options like face enhancement or maximum output quality.
- Busy hours. A full queue adds waiting, not slowness.
Closing the tab does not cancel the job
The swap runs on the server, not in your browser. Your page is only watching. Close it, lose signal, or switch to another app and the frames keep processing.
Signed in, that is completely safe. The finished file waits for you in My Files, with its own progress bar and download button.
Anonymous is where people lose work. Your claim on the job lives in that one browser. Close the tab, switch device, or open a private window, and there is nothing left to link you to the result. Make an account before you start a long clip, not after.
One more trap. Hitting submit twice makes two jobs, and as a free user your second job queues behind your first. You just doubled your own wait.
Why a job can fail at minute seven instead of minute one
Cheap checks run first. File size, clip length, format and a face check on your source photo all fail within seconds. If you got past those, the upload was fine.
Late failures come from the processing itself. There are four common ones.
- The time budget ran out. A video job gets several minutes on the swap server, roughly seven. Past that it is declared failed and your tokens are returned.
- The face disappeared mid clip. A turn, a hand or heavy motion blur can leave frames with nothing to swap.
- The server restarted. A deploy during your job can drop work that was in flight.
- The finishing step failed. After the swap the file still has to be saved, and free results get a watermark pass. That happens at the end, so failures there look like a last second crash.
A failed job is not a charge. The tokens go back automatically, so a retry costs you time and nothing else.
Tell a slow job from a stuck one
Use these four checks in order.
- Still "queued" with a position number? It is waiting, not working. Watch the number drop.
- GIF frame counter moving at all? It is alive. Slow is fine.
- Video under about eight minutes total? Still inside the budget. Leave it.
- Past that with no change? It will be marked failed on its own. Resubmit a shorter clip.
Do not resubmit while the first job is still going. You will be queuing behind yourself.
Set yourself up before a long job
Line up the next five minutes
Test the pipeline with a short clip before you commit to a long one. Three seconds tells you if the source photo works, and it comes back fast. Then run the real clip once, with the settings you already proved.
If you only need a still, use the AI Face Swap photo tool instead. One frame, one pass, and no queue maths to do at all.