Why does apk add ffmpeg fail in the n8n Docker image?

Since n8n 2.1.0 (December 2025) the official image is a hardened Alpine without the apk package manager, so RUN apk add ffmpeg exits with code 127, apk: not found. It is not a typo in your Dockerfile.

Can I run ffmpeg on n8n Cloud?

No. n8n Cloud has no Execute Command node and no shell. The usual video steps, fitting to 9:16, adding a logo, normalising loudness and burning in captions, can be done through an HTTP Request node with a hosted API such as Brand-Pipe.

How do I get ffmpeg into a self-hosted n8n image?

Use a multi-stage Dockerfile: copy /sbin/apk and libapk from a plain alpine stage of the same Alpine version as the n8n tag, then apk add ffmpeg as root and switch back to the node user. The Alpine versions have to match, and the build breaks when n8n bumps its base image.

n8n + ffmpeg

apk add ffmpeg doesn't work in the n8n image anymore

Since n8n 2.1.0 the official Docker image ships without a package manager, so the one-line Dockerfile everyone copied exits with apk: not found. n8n Cloud never had ffmpeg at all. Here is what changed, the Dockerfile that still works for self-hosters, and the no-image route for everyone else.

What changed
Self-hosted: the Dockerfile
n8n Cloud: one HTTP call
Tips for any video API in n8n
Pricing

What changed

The official image has shipped without apk since 2.1.0 (December 2025). It is a hardened Alpine with the package manager removed, not distroless, so the container still has a shell but nothing to install with. The symptom is the same everywhere:

docker build

Step 3/4 : RUN apk add --no-cache ffmpeg
/bin/sh: apk: not found
ERROR: process "/bin/sh -c apk add --no-cache ffmpeg" did not complete successfully: exit code: 127

Two more things landed with 2.0: the Execute Command node is disabled by default in self-hosted instances, and on n8n Cloud it does not exist. So even with ffmpeg in the image you need to re-enable the node before a workflow can call it (the 2.0 migration notes cover the NODES_EXCLUDE setting).

Self-hosted: the Dockerfile that still builds

Copy apk back in from a plain Alpine stage of the same Alpine version as the n8n tag, install, and drop back to the node user:

Dockerfile

FROM alpine:3.24 AS alpine
FROM docker.n8n.io/n8nio/n8n:latest
COPY --from=alpine /sbin/apk /sbin/apk
COPY --from=alpine /usr/lib/libapk.so* /usr/lib/
USER root
RUN apk add --no-cache ffmpeg
USER node

Built on 2026-09-17 against n8n 2.39.6: ffmpeg 8.1.2, n8n still starts as node, the image grows from 270 to 313 MB.
The Alpine version has to match the one inside the n8n image. Check with cat /etc/os-release in a running container before you pick the first line.
It breaks the day n8n bumps its base image. Pin the n8n tag and budget a rebuild for every upgrade.
Then re-enable Execute Command, keep it off the public internet, and write the ffmpeg commands yourself.

Fine for a home lab. Less fine when the video step is the only reason you own a custom image.

n8n Cloud, or no image to maintain: one HTTP call

The step after the video generator is boring and always the same: make the clip 9:16, put the channel logo on it, fix the loudness, maybe burn in captions. Brand-Pipe does exactly that behind one HTTP Request node, so it works on n8n Cloud, on Hostinger, on a locked-down instance, anywhere the node can make a POST.

curl

curl -X POST https://api.brand-pipe.com/v1/process \
  -H "X-API-Key: bp_your_key" \
  -H "Content-Type: application/json" \
  -d '{"video_url": "https://example.com/raw.mp4",
       "logo_url": "https://example.com/logo.png",
       "format": "vertical"}'

The answer is a JSON envelope with a signed download link; the ready-made workflow turns that into binary data for your upload node.

Import n8n-brand-pipe.json (Manual Trigger, Set, HTTP Request, IF, download).
Create a Header Auth credential named Brand-Pipe API Key: header X-API-Key, value your key.
Replace the Set video node with your generator's output and connect the binary output to YouTube, TikTok or storage. n8n-brand-pipe-youtube.json is the scheduled version with a YouTube Shorts upload, shipped disabled so you can test before the OAuth setup.

The loudness part is the one that bites. A 9-second GoPro clip with wind on a covered mic measured -70 LUFS, basically silence. Plain loudnorm pulled it up to -12.3 LUFS, which is 58 dB of hiss. Brand-Pipe measures first and leaves anything below -50 LUFS alone. If you normalise with ffmpeg yourself, add that check.

Get 25 free credits Read the docs

Tips that apply to any video API in n8n

Turn on Never Error in the HTTP Request node options. Otherwise a 4xx throws and your IF node never sees the error message.
Retry On Fail with a paid API can charge you twice when only the response got lost. Send an Idempotency-Key header with {{ $execution.id }}-{{ $itemIndex }}. With just the execution id every item in the run gets the same key.
For longer clips don't keep the request open. Submit a job, put a Wait node on Resume on webhook call with method POST (the default is GET) and send {{ $execution.resumeUrl }} as callback URL. Set a wait limit, or a lost callback parks the execution forever. n8n-brand-pipe-jobs.json is that pattern against POST /v1/jobs.
On the download node set Response Format to File. Autodetect works when the host sends a video content type; not every host does.

Pricing

25 credits free, no card. One credit per video, two if it burns in captions, then €9 for 250 credits. Credits don't expire. Failed renders are refunded automatically (except the rare one that hits the encode time limit). Sync is up to 60 s of output, jobs up to 10 min, 200 MB per source. Renders run in Frankfurt, files are stored in the EU and deleted after 24 h.

Start free All plans

Questions or a case that doesn't fit? support@brand-pipe.com