Picture-in-picture with FFmpeg overlay

ffmpeg -i main.mp4 -i pip.mp4 -filter_complex "[1:v]scale=360:-2[pip];[0:v][pip]overlay=W-w-24:H-h-24[out]" -map "[out]" -map 0:a -c:v libx264 -preset veryfast -crf 20 -c:a copy output.mp4

Scales the second input to 360 px wide, places it 24 px from the bottom-right corner and keeps the audio of the main video untouched. Output: 1280x720, 360 frames, 3.4 MB.

Flags, explained

[1:v]scale=360:-2[pip]

Take the video of the second input, scale it to 360 px wide with an even height, name the result pip.

overlay=W-w-24:H-h-24

Place the inset. Capital W and H are the main frame size, lower-case w and h the inset size.

-map "[out]"

Send the filter graph result to the output. Required once -map is used at all.

-map 0:a

Keep the audio of the first input. -map 1:a takes the second.

pad=iw+8:ih+8:4:4:color=white

Add a 4 px border on each side of the inset.

amix=inputs=2:duration=first

Mix two audio streams, stopping when the first ends.

enable='between(t,2,8)'

Apply the overlay only between 2 s and 8 s.

Variants

ffmpeg -i main.mp4 -i pip.mp4 -filter_complex "[1:v]scale=360:-2[pip];[0:v][pip]overlay=24:24[out]" -map "[out]" -map 0:a -c:v libx264 -preset veryfast -crf 20 -c:a copy output.mp4

ffmpeg -i main.mp4 -i pip.mp4 -filter_complex "[1:v]scale=360:-2,pad=iw+8:ih+8:4:4:color=white[pip];[0:v][pip]overlay=W-w-24:H-h-24[out]" -map "[out]" -map 0:a -c:v libx264 -preset veryfast -crf 20 -c:a copy output.mp4

ffmpeg -i main.mp4 -i pip.mp4 -filter_complex "[1:v]scale=360:-2[pip];[0:v][pip]overlay=W-w-24:H-h-24[out];[0:a][1:a]amix=inputs=2:duration=first[aout]" -map "[out]" -map "[aout]" -c:v libx264 -preset veryfast -crf 20 -c:a aac -b:a 128k output.mp4

ffmpeg -i main.mp4 -i pip.mp4 -filter_complex "[1:v]scale=360:-2[pip];[0:v][pip]overlay=W-w-24:H-h-24:enable='between(t,2,8)'[out]" -map "[out]" -map 1:a -c:v libx264 -preset veryfast -crf 20 -c:a aac -b:a 128k output.mp4

Pitfalls we hit

• Without any -map FFmpeg picks one audio stream by its own rules and does not say which. With two inputs that both carry audio, always map the track you want.
• The overlay ends when the inset ends unless you set eof_action=repeat or pass on the overlay filter, or cut the main video with -shortest.
• Capital W/H are the main video, lower-case w/h the inset. Swapping them puts the inset off-screen.
• scale=360:-1 can give an odd height and fail with yuv420p; -2 keeps it even.
• Video always needs a re-encode here; -c:v copy cannot apply a filter. Only the audio can be copied.

The same edit as a typed operation

create_picture_in_picture_video takes a corner keyword and an inset width, keeps the main audio, and renders the composite in one pass. Inputs are probed and validated before the render, and the result comes back as a signed URL.

{
  "inputs": [
    { "id": "main", "url": "https://example.com/screen.mp4" },
    { "id": "pip", "url": "https://example.com/facecam.mp4" }
  ],
  "options": { "position": "bottom-right", "width": 360 }
}

FAQ