anymotion-agent/anymotion/motion-audio

Audio for Motion Graphics

Sound is roughly half of perceived production value and the cheapest place to gain or lose it. This covers choosing the right track, sourcing it legally, downloading it, and cutting it to picture.

Design for silence first

Most viewers watch muted — in-feed, in an office, on a commute. If the piece only works with sound, it does not work. Sound should elevate, never carry, the message. Build the visual so it reads silent, then add audio as a layer that rewards unmuting.

This is not a stylistic preference; autoplay policies mute by default in every major browser, so silence is the default delivery condition.

Choosing music

Match energy to pacing, not genre to topic. A fintech explainer does not need "corporate" music; it needs music whose rhythmic density matches your cut rate. Fast cuts want a driving pulse. Slow, deliberate camera moves want sustained pads and space.

Find the tempo first. Count your scene changes. If you cut roughly every 3 seconds, you want ~120 BPM (a cut every 6 beats) or ~90 BPM (a cut every 4.5). Cutting on the beat is the single strongest thing you can do to make a film feel professionally edited — and it is free.

The track must have an arc. A loop that never develops makes a 60-second film feel like a 15-second film played four times. Look for tracks with a build, a drop or a lift, and a resolution. Then edit your visual beats to land on the musical ones.

Leave the top end open if there is narration. Music with prominent mid-range vocals or bright leads fights speech. Instrumental, filtered, or pad- heavy tracks sit under voice much better. If the track has vocals, either it is the focus and there is no narration, or it is the wrong track.

Avoid the obvious library sounds. Ukulele-and-whistle, generic uplifting piano, and stock "inspiring corporate" tracks are so overused they actively signal low budget. Slightly unusual instrumentation reads as intentional.

Sound effects

Fewer, better cues. One considered impact beats ten UI blips. Sound everything and it becomes a slot machine.

Sync to the visual accent, not near it. The ear detects audio-visual misalignment at roughly 20–40ms. When in doubt, place the sound 1–2 frames early — audio leading video is far less noticeable than audio lagging, because in the real world light arrives before sound.

Frequency carries meaning. Low frequencies read as weight and arrival; high frequencies as precision and small detail. Match the sound to the size of the thing moving. A large panel sliding in wants a low whoosh with body; a checkbox ticking wants a short, bright click.

Layer for impact. A convincing impact is usually three sounds: a transient (the click), a body (the thump), and a tail (the reverb or decay). Any one alone sounds thin.

Licensing: read this before you download anything

This is where most projects create problems for themselves. The rules:

"Free to download" is not "free to use." A track can be freely downloadable and still prohibit commercial use, require attribution, or forbid modification.

Check the specific licence, not the site's reputation. Sites host tracks under mixed licences — the same platform may carry CC0, CC-BY, and rights-reserved material side by side.

Common licences:

That last row matters: the overwhelming majority of music on YouTube is under the standard licence, which does not grant you any right to extract and reuse it. Downloading it for use in your own video is copyright infringement regardless of the tool used.

Attribution, when required, goes where it can be seen — end card, video description, or project credits. "CC-BY" means the attribution is a condition of the licence, not a courtesy.

Where to source legitimately

Pixabay Music (pixabay.com/music) — the pragmatic first stop. Content is under the Pixabay Content Licence: free for commercial use, no attribution required. Quality is uneven but the catalogue is large and the licence is genuinely permissive. Verify the licence on the individual track page; Pixabay has changed terms before.

YouTube Audio Library (studio.youtube.com → Audio Library) — free tracks and sound effects, clearly labelled as attribution-required or not. Requires a Google account but no channel. Well-tagged by mood, genre, duration, and instrument, which makes searching by feel practical.

Free Music Archive (freemusicarchive.org) — curated,每 track carries an explicit Creative Commons licence. Filter by licence type before browsing so you do not fall for a CC-BY-NC track on a commercial project.

ccMixter (ccmixter.org) — remix-focused, all Creative Commons, strong for electronic and experimental material.

Incompetech (incompetech.com) — Kevin MacLeod's catalogue, CC-BY. Enormously useful and enormously overused; pick the less obvious tracks.

Freesound (freesound.org) — the best source for individual sound effects rather than music. Mixed licences, clearly stated per file. Excellent for whooshes, impacts, and UI sounds.

Zapsplat, Mixkit, Uppbeat — free tiers with varying attribution requirements. Read each one's terms; they differ.

Paid, when the project justifies it: Artlist, Epidemic Sound, Musicbed, Soundstripe. Subscription models with broad clearance, meaningfully better curation, and — importantly — indemnification if a claim arises.

Downloading with yt-dlp

yt-dlp is a general-purpose media downloader. It is a legitimate tool with legitimate uses: retrieving Creative Commons material, content you own, content explicitly licensed for reuse, and archival of your own uploads. Use it for those. Extracting rights-reserved music to put in a video you publish is infringement, and platform Content ID will usually catch it anyway — resulting in a claim, demonetisation, or a takedown.

Install:

pip install -U yt-dlp
# ffmpeg is required for audio extraction and format conversion

Audio only, best quality, as MP3:

yt-dlp -x --audio-format mp3 --audio-quality 0 "URL"

-x extracts audio and discards video. --audio-quality 0 is the best VBR setting.

Keep the original codec (no re-encode, so no generation loss):

yt-dlp -f bestaudio "URL"

Usually yields Opus or AAC in a WebM/M4A container. Prefer this when you will edit further — re-encoding lossy audio twice audibly degrades it.

WAV for editing:

yt-dlp -x --audio-format wav "URL"

Embed metadata and thumbnail:

yt-dlp -x --audio-format mp3 --embed-metadata --embed-thumbnail "URL"

Check the licence before downloading:

yt-dlp --print "%(license)s | %(title)s | %(uploader)s" "URL"

If this prints Creative Commons Attribution licence (reuse allowed), you are on solid ground and you owe attribution. If it prints nothing useful or Standard YouTube License, do not use the track — find an equivalent on Pixabay or the Audio Library instead.

Search YouTube's CC-licensed material by adding ,creativecommons to the search filter on the site, then verify each result with the --print command above.

Download a whole playlist's audio:

yt-dlp -x --audio-format mp3 -o "%(playlist_index)s - %(title)s.%(ext)s" "PLAYLIST_URL"

Rate-limit to be a good citizen:

yt-dlp --sleep-interval 3 --max-sleep-interval 8 -x "URL"

Editing with ffmpeg

ffmpeg handles everything you need to fit a track to picture.

Trim to a section (from 0:15, lasting 30s):

ffmpeg -ss 15 -t 30 -i input.mp3 -c copy output.mp3

Putting -ss before -i seeks fast; -c copy avoids re-encoding.

Fade in and out (2s fade in, 3s fade out ending at 30s):

ffmpeg -i input.mp3 -af "afade=t=in:st=0:d=2,afade=t=out:st=27:d=3" output.mp3

A fade-out is almost always necessary — music that stops abruptly at the end of a film sounds like an error.

Normalize loudness to broadcast standard:

ffmpeg -i input.mp3 -af loudnorm=I=-16:TP=-1.5:LRA=11 output.mp3

-16 LUFS is the streaming/web target. -14 is Spotify's; -23 is EBU R128 broadcast. Consistent loudness across cues is what separates a mixed film from a pile of clips.

Duck music under narration (sidechain compression):

ffmpeg -i music.mp3 -i voice.mp3 -filter_complex \
  "[0:a][1:a]sidechaincompress=threshold=0.03:ratio=8:attack=20:release=400[out]" \
  -map "[out]" ducked.mp3

The music automatically drops whenever the voice is present and returns when it stops. Far better than a static volume reduction.

Mix music and effects:

ffmpeg -i music.mp3 -i sfx.wav -filter_complex \
  "[0:a]volume=0.4[a0];[a0][1:a]amix=inputs=2:duration=longest" mixed.mp3

Change tempo without changing pitch (fit a track to a fixed duration):

ffmpeg -i input.mp3 -af "atempo=1.08" output.mp3

atempo accepts 0.5–2.0; chain two filters for larger changes. Beyond about ±10% it becomes audible.

Extract audio from a video file:

ffmpeg -i video.mp4 -vn -acodec copy audio.aac

Convert for web delivery:

# AAC — universal support
ffmpeg -i input.wav -c:a aac -b:a 192k output.m4a
 
# Opus — better quality per byte, modern browsers
ffmpeg -i input.wav -c:a libopus -b:a 128k output.opus

Procedural audio, when files are a problem

If you cannot resolve licensing, need zero network requests, or want sound that scales with a parameter, synthesize it. The Web Audio API generates clicks, whooshes, and impacts from oscillators and noise with no assets at all:

function click(ctx, when = ctx.currentTime) {
  const osc = ctx.createOscillator();
  const gain = ctx.createGain();
  osc.frequency.setValueAtTime(2400, when);
  osc.frequency.exponentialRampToValueAtTime(800, when + 0.03);
  gain.gain.setValueAtTime(0.3, when);
  gain.gain.exponentialRampToValueAtTime(0.001, when + 0.06);
  osc.connect(gain).connect(ctx.destination);
  osc.start(when);
  osc.stop(when + 0.08);
}

This is fully licence-free, adds no bytes, and is deterministic — the same code produces the same sound every time. It will not replace music, but it covers UI sounds completely.

Practical checklist

1. Does the piece communicate with sound off? If not, fix the visual first.
2. Is the licence explicitly stated, and does it permit your use? Save a copy of

the licence text alongside the file.

1. Does the tempo match your cut rate? Count it before committing.
2. Are your key visual beats landing on musical beats?
3. Is the music ducked under any narration?
4. Is everything normalized to a consistent loudness (−16 LUFS for web)?
5. Does the track fade out rather than stop dead?
6. If attribution is required, is it actually in the deliverable?

Common mistakes