fakestream

Synthetic test video for people building AV players, generated from nothing and served over HTTP.

Testing a player against real providers is slow and incomplete. Accounts and contracts gate the streams, the awkward cases only appear when the schedule happens to produce them, and nothing about the content is under your control. Sample files scattered around the internet cover the gaps badly, with unclear licensing and no way to know what a file is supposed to contain. fakestream replaces both. One binary serves every stream shape a real provider might send, on localhost, with content designed so that player bugs identify themselves:

• A beep sounds on every second boundary and the picture flashes a white marker on the same frame, so drift between audio and video is visible and audible at once.
• Captions and subtitles are numbered, so a dropped or mistimed cue is a fact rather than a feeling.
• Live pictures carry a UTC clock, elapsed time and a frame counter, so end to end latency can be measured by holding a real clock next to the screen, and a stalled or looping player is obvious.
• Caption channels and subtitle languages carry visibly different text, so a player rendering the wrong track betrays itself immediately.

Every output is made from nothing. There are no seed files and no third-party sample clips, so there is nothing to license and nothing to download beyond the tool itself. Caption text is generated lorem ipsum.

Install

Each release ships a self-contained binary per platform, with ffmpeg and x264 built in. Nothing else needs installing.

macOS (Apple Silicon) and Linux (x86_64)

curl -fsSL https://raw.githubusercontent.com/ektotv/fakestream/main/scripts/install.sh | sh

The script detects the platform, downloads the latest release, checks it against the release's SHA256SUMS, and unpacks it into the current directory. It never writes outside that directory; putting the binary on your PATH is a separate step it prints and leaves to you:

sudo install fakestream-v*/fakestream /usr/local/bin/

Windows (x86_64)

curl.exe -fsSL -o install.bat https://raw.githubusercontent.com/ektotv/fakestream/main/scripts/install.bat
.\install.bat

Two lines rather than one joined with &&, which Windows PowerShell does not accept, and curl.exe rather than curl, which PowerShell aliases to something that takes different flags. Written this way it pastes into cmd and any PowerShell alike.

Manually

Download the archive for your platform from the latest release, check it against the release's SHA256SUMS, unpack it, and run the binary inside.

On macOS the binary is not signed or notarised. Downloads made with a browser carry a quarantine flag that blocks it with "cannot be opened"; either allow it under System Settings, Privacy & Security, or clear the flag:

xattr -d com.apple.quarantine fakestream

Downloads made with curl, including the install script above, carry no flag and run as they are.

Whichever way it arrived, every binary reports exactly what it is:

$ fakestream --version
fakestream 0.1.0 (986e444, built 2026-08-03)

Serving streams

fakestream serve

Running fakestream with no command does the same. Open http://localhost:8080 for the list of streams with their purposes, and point a player at any of them.

VOD fixtures are generated the first time they are requested and cached afterwards, so the first play of one takes a few seconds and every later play is immediate. Live streams are produced in real time as they are watched. The progressive stream ends when its viewer disconnects; the live HLS writer is shared by every viewer and stops about a minute after the last request for anything in its playlist.

Every request is logged with a UTC timestamp, the client address and the response, which turns the server into a diagnostic tool in itself; a player that stops polling a live playlist, or fetches segments it should have cached, shows up in the log.

12:34:56.789Z  127.0.0.1:60447  GET 200 /live/hls/stream0.m3u8 2ms

Generating up front

fakestream build

Generates every VOD fixture into the cache without serving, which takes a couple of minutes. Useful ahead of a demo or on a machine that will serve offline. Live streams have nothing to build, they are produced as watched.

Options

Both commands take the same flags:

flag
--dir PATH	where fixtures are cached, default ./fixtures
--port PORT	port to listen on, default 8080 (serve only)
--quiet	drop the progress bar, keeping one line per fixture
--verbose	let ffmpeg log everything, for diagnosing a bad file
--licenses	print the licences of everything in the binary
-v, --version	print the version, commit and build date
-h, --help	print usage

What it produces

path	what it tests
vod/basic.mp4	playback, seeking and AV sync, with no captions
vod/cea608.ts	CEA-608 hidden in the video, announced by nothing
vod/cea608-dual.ts	choosing between caption channels CC1 and CC2
vod/cea708.ts	CEA-708 service data alongside 608, as a broadcast carries both
vod/cea-pmt.ts	the same 608 and 708, this time announced in the PMT by a caption_service_descriptor
vod/dvbsub.ts	DVB bitmap subtitles on their own track
vod/tx3g.mp4	tx3g timed text, the usual subtitles inside MP4
vod/ttml.mp4	TTML as stpp, what DASH and CMAF use
vod/subrip.mkv	SubRip, plain text with no styling
vod/ass.mkv	ASS, which carries styling and positioning
vod/webvtt.mkv	WebVTT away from HLS
vod/multilingual.mkv	four subtitle tracks, tagged eng, fra, spa and jpn
hls/ts/master.m3u8	HLS with MPEG-TS segments and a WebVTT rendition
hls/fmp4/master.m3u8	HLS with fragmented MP4 segments
live/stream.ts	endless progressive MPEG-TS, the classic IPTV shape, no captions
live/hls/master.m3u8	live HLS, a rolling window that cannot be seeked before
live/cea608.ts	endless live CEA-608, each caption stamped with the time it is due
live/cea708.ts	endless live CEA-708 alongside 608 in the same SEI
live/cea608-pmt.ts	live CEA-608, announced in every PMT
live/cea708-pmt.ts	live CEA-608 and 708, announced in every PMT
live/hls-cea608/master.m3u8	live HLS carrying CEA-608 across segment boundaries
live/hls-cea708/master.m3u8	live HLS carrying CEA-708 across segment boundaries

One gap is known and deliberate. HLS carries one subtitle language, which is a limit of ffmpeg's HLS muxer rather than a choice, and the multilingual Matroska fixture covers language selection meanwhile.

Building from source

For a binary that carries everything with it and needs nothing installed:

./scripts/build-ffmpeg.sh
FFMPEG_PKG_CONFIG_PATH=$PWD/third_party/ffmpeg/lib/pkgconfig \
  cargo build --release --no-default-features

The first step builds a static ffmpeg and x264 into third_party/ffmpeg, which takes about a minute and is only done once. The binary lands at target/release/fakestream and links nothing beyond the operating system's own libraries.

For development, linking the ffmpeg already on the machine is quicker:

cargo build

That needs ffmpeg 8 installed.

Licence

GPL-3.0. The binary carries the licence texts of everything statically linked inside it, ffmpeg, x264, libcaption, the Noto Sans font and every Rust crate; fakestream --licenses prints the lot.