Scale a copy to cover 1080x1920, blur it with gblur at sigma 40 and overlay the original at full width: 16:9 becomes 9:16 with no black bars. ReelSDK's free vertical video converter does it in your browser, or the API at $0.02 per compute minute.
Blurred background video with ffmpeg
The 1920x1080 frame shrinks to a sharp 1080x608 band in the middle, with 656 px of soft colour above and below it. ReelSDK's slideshow engine frames photos the same way, with a blurred backdrop drawn at 1/8 size.
GuidesUpdated
ffmpeg -i in.mp4 -filter_complex \
"[0:v]scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,gblur=sigma=40[bg];[0:v]scale=1080:-2[fg];[bg][fg]overlay=(W-w)/2:(H-h)/2" \
-c:v libx264 -crf 20 -preset veryfast -c:a copy -movflags +faststart vertical.mp4[0:v] feeds two chains. The first scales a copy until it covers 1080x1920, crops the overflow and blurs it; the second scales the original to 1080 px wide. overlay centres the sharp layer on the blur. On a 10-second input it wrote a file of about 4 MB, with the audio copied.
Variations
The other commands you will need
Darker backdrop from a small copy
Blur a 136x240 copy with sigma 6, multiply its colours by 0.55 and scale it back up. That is the backdrop our slideshow engine draws behind photos. To a null output it was no faster than the full-size blur, so pick it for the darker look, not for speed.
small dark backdrop ffmpeg -i in.mp4 -filter_complex \ "[0:v]scale=136:240:force_original_aspect_ratio=increase,crop=136:240,gblur=sigma=6,colorchannelmixer=rr=0.55:gg=0.55:bb=0.55,scale=1080:1920[bg];[0:v]scale=1080:-2[fg];[bg][fg]overlay=(W-w)/2:(H-h)/2" \ -c:v libx264 -crf 20 -preset veryfast -c:a copy -movflags +faststart vertical-small.mp4Vertical clip into a 16:9 frame
The same graph the other way round: the blurred copy covers 1920x1080 and the original is scaled to 1080 px tall. Our 1080x1920 clip came out at 1920x1080, with the blur filling the sides.
vertical to horizontal ffmpeg -i clip.mp4 -filter_complex \ "[0:v]scale=1920:1080:force_original_aspect_ratio=increase,crop=1920:1080,gblur=sigma=40[bg];[0:v]scale=-2:1080[fg];[bg][fg]overlay=(W-w)/2:(H-h)/2" \ -c:v libx264 -crf 20 -preset veryfast -c:a copy -movflags +faststart horizontal.mp4
The flags
The filters in this graph
| Flag | What it does | Used here |
|---|---|---|
gblur=sigma=40 | Gaussian blur; sigma sets the strength, from 0 to 1024, default 0.5 | Backdrop |
crop=1080:1920 | Keeps the centre of the oversized copy | Backdrop |
scale=1080:-2 | Fixes the width and picks the nearest even height that keeps the ratio | Front layer, 1080x608 |
force_original_aspect_ratio=increase | Scales until both sides cover the target size, so the crop leaves no gaps | Backdrop |
overlay=(W-w)/2:(H-h)/2 | Centres the second input on the first; W and H are the background's size | Both layers |
colorchannelmixer=rr=0.55:gg=0.55:bb=0.55 | Multiplies red, green and blue to darken | Small backdrop |
boxblur=40:2 | A cheaper box blur; its file came out almost the same size as gblur's | Not used |
-c:a copy | Copies the audio track without re-encoding it | All three |
Sources: ffmpeg filters documentation, ReelSDK slideshow engine, checked November 2026.
Why cover first, then blur
A 16:9 frame on a 9:16 canvas leaves two empty bands. You can fill them with black, as the pad filter does, or with the picture itself. force_original_aspect_ratio=increase grows the copy until no edge is short, and crop keeps the middle. Blown up that far, the copy is soft and blocky, which is why it gets a heavy blur: at sigma 40 only colour and light are left.
The front layer uses scale=1080:-2. The -2 asks for an even height, and H.264 in yuv420p needs even sizes. Here it lands at 1080x608, centred at y=656.
Blur strength and cost
gblur's sigma runs from 0 to 1024 and defaults to 0.5, which barely shows. On a 1080-wide backdrop, sigma 40 leaves only colour and light, as the poster above shows.
Blurring a small copy and scaling it up looks much the same and adds a darker tone. On our machine it was not faster to a null output, so choose by look.
Photos instead of video
For still photos you don't need this graph. The slideshow API has fit: "contain", which floats each photo over a blurred, darkened copy of itself: a backdrop at 1/8 size, sigma 6 for 1080x1920 and colours times 0.55, the same recipe as the second command.
For a video file, send the graph as it is to the ffmpeg endpoint. Each job is billed by compute time; see pricing.
Without installing ffmpeg
The same command as one API call
Send the same args to POST /v1/ffmpeg with the clip's URL in inputs. Each job runs ffmpeg 8.1 on a fresh machine at $0.02 per compute minute, billed per second; this one finished in about 8 seconds, so it cost under a third of a cent.
curl "https://api.reelsdk.com/v1/ffmpeg?wait=60" \
-H "Authorization: Bearer $REELSDK_KEY" \
--json '{
"inputs": {
"in.mp4": "https://api.reelsdk.com/f/lib/demo/landscape.mp4"
},
"args": [
"-i", "in.mp4", "-filter_complex",
"[0:v]scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,gblur=sigma=40[bg];[0:v]scale=1080:-2[fg];[bg][fg]overlay=(W-w)/2:(H-h)/2",
"-c:v", "libx264", "-crf", "20", "-preset", "veryfast", "-c:a", "copy",
"-movflags", "+faststart", "vertical.mp4"
],
"outputs": ["vertical.mp4"]
}'FAQ
Questions, answered
How do I turn a 16:9 video into 9:16 without black bars?
Scale a copy to cover 1080x1920, crop it, blur it with gblur, then overlay the original scaled to 1080 px wide in the centre. On a 1920x1080 clip that writes a 1080x1920 file with every frame and the audio copied.
What blur strength should I use?
gblur defaults to sigma 0.5, which barely shows. Sigma 40 on a 1080-wide backdrop leaves only colour and light; a 136x240 copy blurred at sigma 6 and scaled up gives a similar look.
Why does the filter use [0:v] twice?
Each chain needs its own copy of the input. Naming [0:v] twice feeds both; split=2 does the same inside the graph.
Can I run this without installing ffmpeg?
Yes. Send the same args to POST /v1/ffmpeg with the video's URL in inputs. It runs ffmpeg 8.1 on a fresh machine per job for $0.02 per compute minute, billed per second with a 1 s minimum.
- Real estate listing videosTurn landscape listing photos into a vertical reel with a voiceover, music and captions, without cropping a single photo.
- Vertical videoTurn 16:9 footage into a 1080x1920 vertical video with a blurred fill in one ffmpeg call, about a quarter cent per 30 s clip.
- Vertical video converterTurn a landscape video into a vertical one in your browser, with a blurred background, free and without uploading it.
- Letterbox and pad a video with ffmpegFit any video into a 9:16, 16:9 or square frame with black or coloured bars, then find and strip bars with cropdetect.
- Crop a video with ffmpegCut a frame down with crop=w:h:x:y: a centered 9:16 or square crop, a fixed region, and cropdetect to find black bars.
- Resize a video with ffmpegScale a video with scale=w:h: keep the aspect ratio with -2, fit a 1080x1920 frame with pad, and fix the odd-size error.
- ffmpeg -filter_complex explainedLabels, commas and semicolons: how a filtergraph wires inputs to outputs, with four working graphs and the errors they fix.
- height not divisible by 2libx264 refuses a 4:2:0 frame with an odd width or height. Scale with -2, round odd sources to even, or send the args to an API.
- ReelSDK vs self-hosted FFmpegTranscoding a minute of 1080p to 720p takes about 24 seconds of compute: about 0.8 cents on ReelSDK, about a tenth of a cent on a busy 4 vCPU box. Below about 11,000 minutes a month, the API costs less than an always-on box.
Sources, checked November 2026
Ready to build?
Ship video from your app today
Your first $0.50 is free, and one key covers every endpoint.
