Ffmpeg Transcode

Ffmpeg Transcode

Context: $ARGUMENTS

Quick start

• Web MP4 (H.264 + AAC, most compatible): → Step 3, recipe A
• Change container only, keep codecs: → Step 1 (remux path)
• Smaller file at same quality (HEVC/AV1): → Step 3, recipes B / C

When to use

• User says "convert", "re-encode", "transcode", "change format", "make smaller".
• Target codec differs from source, OR target container doesn't support source codec.
• CRF / bitrate / 2-pass / preset tuning for size vs quality.
• For GPU encoding use ffmpeg-hwaccel; for trimming/joining use ffmpeg-cut-concat; for filters/scaling use ffmpeg-video-filter; for raw inspection use ffmpeg-probe.

Step 1 — Pick container + codecs (and decide remux vs re-encode)

First probe the input:

ffprobe -v error -show_entries stream=index,codec_type,codec_name,pix_fmt,profile,channels,sample_rate -of json "$IN"

Decision tree:

1. If the user only wants a different container AND the source codecs are supported by the target container (see matrix below), remux — no quality loss, fast:

   ffmpeg -i in.mkv -c copy -map 0 out.mp4
   

   Caveats: H.264/HEVC in raw .ts vs .mp4 need a bitstream filter — see ffmpeg-bitstream. AAC from .ts → .mp4 needs aac_adtstoasc.
2. If codecs are incompatible with the target container (e.g. Opus → MP4 is technically allowed but fragile; FLAC → MP4 not allowed), re-encode that stream only.
3. Otherwise full re-encode — continue to Step 2.

Container → codec compatibility matrix:

If unsure which codec to pick, default to H.264 + AAC in MP4 — works on every browser, every phone, every social platform.

Step 2 — Choose rate control

Three modes, pick exactly one per video encode:

• CRF (constant quality) — best default. Pick a target quality; filesize varies with content. Use when output size doesn't have to be exact.
• Bitrate target — -b:v 5M. Single pass is fine for live/streaming targets; otherwise 2-pass is higher quality at the same size.
• 2-pass — use when you must hit a specific size (e.g. Twitter 512MB, Discord 25MB). Two ffmpeg invocations; first pass writes stats to ffmpeg2pass-*.log.

CRF typical ranges (NOT comparable across codecs):

2-pass recipe (size-targeted):

# Pass 1: analyse only, no audio, discard output
ffmpeg -y -i "$IN" -c:v libx264 -b:v 5M -preset medium \
       -pass 1 -an -f null /dev/null

# Pass 2: real encode, reuse stats
ffmpeg -i "$IN" -c:v libx264 -b:v 5M -preset medium \
       -pass 2 -c:a aac -b:a 128k \
       -movflags +faststart "$OUT"

Compute -b:v from a size budget: bitrate_bps = target_size_bits / duration_seconds − audio_bps.

Step 3 — Build the command

Recipe A — Web MP4 (default go-to):

ffmpeg -i "$IN" \
  -c:v libx264 -preset medium -crf 20 \
  -pix_fmt yuv420p \
  -c:a aac -b:a 128k -ac 2 \
  -movflags +faststart \
  "$OUT.mp4"

Recipe B — HEVC in MKV (archive / smaller size):

ffmpeg -i "$IN" \
  -c:v libx265 -preset medium -crf 22 \
  -pix_fmt yuv420p10le \
  -c:a libopus -b:a 128k \
  "$OUT.mkv"

Put HEVC in MP4 instead by adding -tag:v hvc1 (Apple/QuickTime requires it).

Recipe C — AV1 in MP4 (modern web / YouTube-style):

ffmpeg -i "$IN" \
  -c:v libsvtav1 -preset 6 -crf 30 \
  -pix_fmt yuv420p10le \
  -c:a libopus -b:a 96k \
  -movflags +faststart \
  "$OUT.mp4"

(Use -preset 4 for better quality / slower, -preset 8 for faster / larger.)

Recipe D — WebM (VP9 + Opus):

ffmpeg -i "$IN" \
  -c:v libvpx-vp9 -b:v 0 -crf 32 -row-mt 1 -cpu-used 2 \
  -pix_fmt yuv420p \
  -c:a libopus -b:a 96k \
  "$OUT.webm"

Recipe E — ProRes for editing (MOV):

ffmpeg -i "$IN" \
  -c:v prores_ks -profile:v 3 -pix_fmt yuv422p10le \
  -c:a pcm_s16le \
  "$OUT.mov"

Profiles: 0=proxy, 1=lt, 2=standard, 3=hq, 4=4444, 5=4444xq.

Stream selection (-map):

# Keep video 0 + audio 1 only, drop all subs:
ffmpeg -i "$IN" -map 0:v:0 -map 0:a:1 -c:v libx264 -crf 20 -c:a aac "$OUT"
# Preserve all streams + copy metadata + copy chapters:
ffmpeg -i "$IN" -map 0 -map_metadata 0 -map_chapters 0 -c copy "$OUT"

Preset runner:

uv run ${CLAUDE_SKILL_DIR}/scripts/transcode.py \
  --input "$IN" --output "$OUT" --preset web-mp4

Presets: web-mp4 (H.264/AAC MP4), hevc-mkv, web-webm, av1-mp4, prores, archive.

Step 4 — Verify output

ffprobe -v error -show_format -show_streams -of json "$OUT" | jq '.format.duration, .streams[].codec_name'

Check:

• Duration roughly matches input (within ~0.1s, short container padding is normal).
• Video codec_name + audio codec_name are what you asked for.
• pix_fmt is yuv420p if you need browser/phone playback.
• For MP4: ffprobe -v trace ... 2>&1 | head | grep moov should show moov atom early (or simply: the file starts playing instantly in a browser = faststart worked).

Gotchas

• yuv420p is required for broad playback (browsers, iOS, most TVs). ProRes captures and RGB sources often land on yuv444p/yuv422p — always add -pix_fmt yuv420p for web output, or yuv420p10le for 10-bit HDR-ish HEVC/AV1.
• -movflags +faststart is MP4/MOV only. Does nothing for MKV/WebM. Without it, web players have to download the whole file before playback starts.
• Stream copy across containers may need a bitstream filter (e.g. H.264 MP4 → TS, AAC TS → MP4). See the ffmpeg-bitstream skill.
• CRF values are NOT comparable across codecs. x264 crf 23 ≠ x265 crf 23 ≠ av1 crf 23. Use the table in Step 2.
• -crf 0 is lossless only for libx264 / libx265. For AV1 / VP9 it's the best-quality CRF value but still lossy. For true lossless VP9 use -lossless 1.
• -b:v is ignored when -crf is set on libx264/libx265 unless you also pass -maxrate/-bufsize (then CRF becomes a capped-CRF ceiling). Pick one mode.
• Audio must be re-encoded when container doesn't support source codec — e.g. Opus in MP4 is shaky, FLAC in MP4 is unsupported, AAC in WebM is unsupported. Re-encode the audio even if you're copying video: -c:v copy -c:a aac -b:a 128k.
• HEVC in MP4 needs -tag:v hvc1 to play on Apple QuickTime / Safari / iOS.
• 2-pass requires matching -b:v and -preset between the two passes. Stats log is per-working-directory — run both passes from the same cwd.
• libfdk_aac is non-free and absent from most ffmpeg builds. Use the native aac encoder unless you know your build has fdk.

Examples

Example 1: "Make this MOV playable in browsers"

ffmpeg -i in.mov -c:v libx264 -preset medium -crf 20 -pix_fmt yuv420p \
       -c:a aac -b:a 128k -movflags +faststart out.mp4

Example 2: "Shrink a 4GB recording for Discord (25MB cap)"

Duration 10 min = 600s. Budget: 25 MB × 8 / 600s ≈ 333 kbps total. Reserve 64k audio, leave 260k video. 2-pass mandatory at that ratio:

ffmpeg -y -i in.mp4 -c:v libx264 -b:v 260k -preset slow -pass 1 -an -f null /dev/null
ffmpeg    -i in.mp4 -c:v libx264 -b:v 260k -preset slow -pass 2 \
          -c:a aac -b:a 64k -ac 1 -movflags +faststart out.mp4

Example 3: "Change MKV container to MP4 without re-encoding"

ffprobe in.mkv            # confirm codecs are H.264/HEVC + AAC/AC3
ffmpeg -i in.mkv -c copy -map 0 -movflags +faststart out.mp4

If subtitle streams are SRT/ASS, add -map 0:v -map 0:a to drop them (MP4 only takes mov_text) or add -c:s mov_text to convert.

Example 4: "Archive master to HEVC 10-bit"

ffmpeg -i master.mov -c:v libx265 -preset slow -crf 18 -pix_fmt yuv420p10le \
       -x265-params "profile=main10" \
       -c:a flac -compression_level 8 archive.mkv

Troubleshooting

Error: Could not find tag for codec opus in stream #1, codec not currently supported in container

Cause: Opus audio in MP4 isn't universally supported by ffmpeg's MP4 muxer. Solution: Use MKV/WebM, or re-encode audio to AAC: -c:a aac -b:a 128k.

Error: [libx264 @ ...] height not divisible by 2

Cause: libx264 with yuv420p requires even dimensions. Solution: Add -vf "scale=trunc(iw/2)*2:trunc(ih/2)*2" or an explicit even target.

Output plays but browsers won't start until fully downloaded

Cause: moov atom at end of MP4. Solution: Add -movflags +faststart. (MP4/MOV only.)

Output has green / purple tint on iOS

Cause: yuv444p or yuv422p pixel format. Solution: Add -pix_fmt yuv420p.

HEVC MP4 won't play on Apple devices

Cause: tag is hev1 instead of hvc1. Solution: Re-mux with ffmpeg -i in.mp4 -c copy -tag:v hvc1 out.mp4.

"Unknown encoder 'libfdk_aac'"

Cause: build doesn't include non-free fdk. Solution: Swap libfdk_aac → aac (native encoder is fine for ≥128 kbps stereo).

Reference docs

• Deep per-encoder options (presets, tunes, profiles, levels, AV1/VP9 tile options, Opus application modes, pixel-format & color cheat sheet) → references/codecs.md.