Blurred Background

Blurred background
vertical video in ffmpeg.

The full filter graph, what every single part of it does, how to batch it across a folder, and the three mistakes that produce broken output.

See the full format spec →Find a clip

The command

Complete filter graph

ffmpeg -y -i input.mp4 -filter_complex "\
  [0:v]scale=1080:1920:force_original_aspect_ratio=increase,\
       crop=1080:1920,boxblur=40:4,setsar=1[bg];\
  [0:v]scale=1080:-2[fg];\
  [bg][fg]overlay=(W-w)/2:(H-h)/2:format=auto,format=yuv420p[v]" \
  -map "[v]" -map 0:a? -c:v libx264 -preset veryfast -crf 20 \
  -c:a copy -movflags +faststart output.mp4

Breakdown

Every part, explained

Fragment	What it does
`[0:v]`	The video stream of input 0. Used twice — once for background, once for foreground.
`force_original_aspect_ratio=increase`	Scales up until the frame covers the canvas, overflowing on one axis rather than fitting inside it.
`crop=1080:1920`	Trims that overflow to exact canvas size, producing a full-bleed background.
`boxblur=40:4`	Radius 40, 4 passes. Repeated box blur approximates gaussian at a fraction of the cost.
`setsar=1`	Forces square pixels. Without it, mismatched sample aspect ratios between layers break the overlay.
`scale=1080:-2`	Foreground at full width; height computed automatically and rounded even for H.264.
`overlay=(W-w)/2:(H-h)/2`	Centres foreground over background on both axes using canvas and layer dimensions.
`format=yuv420p`	Mandatory for social platforms. Without it ffmpeg may emit yuv444p, which most uploaders reject.
`-map 0:a?`	Includes audio only if present. The `?` prevents a silent source from failing the job.
`-movflags +faststart`	Moves the MP4 index to the front so the file streams before it finishes downloading.

Batch

Processing a whole folder

macOS / Linux

mkdir -p out
for f in *.mp4; do
  ffmpeg -y -i "$f" -filter_complex "\
    [0:v]scale=1080:1920:force_original_aspect_ratio=increase,\
         crop=1080:1920,boxblur=40:4,setsar=1[bg];\
    [0:v]scale=1080:-2[fg];\
    [bg][fg]overlay=(W-w)/2:(H-h)/2:format=auto,format=yuv420p[v]" \
    -map "[v]" -map 0:a? -c:v libx264 -preset veryfast -crf 20 \
    -c:a copy -movflags +faststart "out/${f%.mp4}-vertical.mp4"
done

Windows (PowerShell)

mkdir out -Force
Get-ChildItem *.mp4 | ForEach-Object {
  ffmpeg -y -i $_.Name -filter_complex "[0:v]scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,boxblur=40:4,setsar=1[bg];[0:v]scale=1080:-2[fg];[bg][fg]overlay=(W-w)/2:(H-h)/2:format=auto,format=yuv420p[v]" -map "[v]" -map 0:a? -c:v libx264 -preset veryfast -crf 20 -c:a copy -movflags +faststart "out/$($_.BaseName)-vertical.mp4"
}

Failure modes

Three mistakes that break output

Missing yuv420p

The file encodes fine and plays locally, then fails to upload with an unhelpful error. Always end the chain with format=yuv420p.

Hardcoded height

Works for exactly-16:9 sources, silently stretches everything else. Use -2 and let ffmpeg compute it.

Missing setsar=1

Sources with non-square pixels produce a distorted or failed overlay. Cheap to add, annoying to debug.

Performance

Speeding it up

The blur is the expensive filter. If you're processing at volume:

Blur a smaller layer. Scale down, blur, then scale back up — visually near-identical, meaningfully cheaper, since the result is heavily blurred anyway.
Keep -preset veryfast. The blurred region conceals most encoding artefacts, so slower presets buy very little here.
Don't touch the audio. -c:a copy skips audio re-encoding entirely.

# cheaper background: downscale → blur → upscale
[0:v]scale=270:480,boxblur=10:4,scale=1080:1920,setsar=1[bg];

FAQ

Common questions

What does boxblur=40:4 mean?

Radius 40 pixels, applied over 4 passes. Multiple passes of box blur approximate a gaussian blur at much lower computational cost.

Why use -2 instead of a fixed height?

It tells ffmpeg to compute the height from the width while preserving aspect ratio and rounding to an even number, which H.264 requires. A fixed height distorts any source that isn't exactly 16:9.

Can I use GPU acceleration?

Yes — h264_nvenc, h264_qsv, or h264_videotoolbox can replace libx264 for encoding. The blur filter itself still runs on CPU unless you use the CUDA filter chain.

How do I make the blur stronger?

Increase the boxblur radius. Values between 30 and 50 work well; beyond that the background loses all structure and stops reading as related to the foreground.