Skip to content

ffmpeg.md

Latest commit

 

History

History
122 lines (100 loc) · 5.48 KB

ffmpeg.md

File metadata and controls

122 lines (100 loc) · 5.48 KB
Raw

FFmpeg Input Recipes

The publisher's preferred fast path is already fragmented MP4 input. You can generate that with ffmpeg by copying compatible AAC-LC, H.264, or HEVC streams and enabling CMAF-style fragmentation flags.

Fragment an Existing MP4

ffmpeg -i input.mp4 \
  -map 0:v -map 0:a \
  -map_metadata -1 \
  -sn -dn \
  -c:v copy \
  -c:a copy \
  -movflags +frag_keyframe+empty_moov+default_base_moof+separate_moof \
  -f mp4 fragmented.mp4

Re-Encode to Compatible H.264 or HEVC

If the source codecs are not already compatible, re-encode instead of copying.

H.264:

ffmpeg -i bbb_sunflower_2160p_60fps_normal.mp4 \
  -map 0:v:0 -map 0:a:0 \
  -map_metadata -1 \
  -sn -dn \
  -c:v libx264 -preset medium -r 30 -g 60 -keyint_min 60 -sc_threshold 0 -bf 0 \
  -c:a aac -b:a 160k -ar 48000 -ac 2 \
  -movflags +frag_keyframe+empty_moov+default_base_moof+separate_moof \
  -f mp4 sunflower-frag.mp4

HEVC:

ffmpeg -i bbb_sunflower_2160p_60fps_normal.mp4 \
  -map 0:v:0 -map 0:a:0 \
  -map_metadata -1 \
  -sn -dn \
  -c:v libx265 -preset medium -r 30 -g 60 -keyint_min 60 -sc_threshold 0 -bf 0 \
  -c:a aac -b:a 160k -ar 48000 -ac 2 \
  -movflags +frag_keyframe+empty_moov+default_base_moof+separate_moof \
  -f mp4 sunflower265-frag.mp4

Practical Notes

  • -map 0:v -map 0:a keeps only video and audio streams, excluding subtitle and other non-A/V tracks
  • -map 0:v:0 -map 0:a:0 uses the first audio stream if multiple exist
  • -sn -dn explicitly disables subtitle and data or text streams
  • -map_metadata -1 drops container-level metadata from the output
  • +frag_keyframe starts a new fragment on keyframes
  • +empty_moov writes initialization metadata up front
  • +default_base_moof and +separate_moof produce a layout that is easier for fragmented-MP4 pipelines to consume
  • omit +separate_moof only if you are sure downstream tooling can parse interleaved multi-track fragments
  • when audio appears in the catalog but no audio media objects are sent, regenerate with +separate_moof
  • for HEVC, prefer streams that are already hvc1-compatible
  • if a source is tagged hev1 but keeps VPS/SPS/PPS only in the init segment, the publisher normalizes the advertised codec and emitted init segment to hvc1
  • if HEVC samples include in-band parameter sets, the publisher preserves hev1
  • if you start from a progressive MP4, this project can remux it internally, but pre-fragmented input is simpler and more efficient

Live Fragmented MP4 Stdin Publishing

For live encoder pipelines, the publisher can consume fragmented MP4 directly from standard input.

This live path expects ffmpeg to emit track-separated fragments, where each moof + mdat pair belongs to a single media track. Use +separate_moof when generating the stream. Without +separate_moof, audio and video may be carried inside the same moof, which is not the intended input layout for the current live parser.

ffmpeg -stream_loop -1 -re -i bbb_sunflower_1080p_30fps_normal.mp4 \
  -map 0:v:0 -map 0:a:0 \
  -c:v libx264 -preset medium -r 30 -g 60 -keyint_min 60 -sc_threshold 0 -bf 0 \
  -c:a aac -b:a 160k -ar 48000 -ac 2 \
  -movflags +frag_keyframe+empty_moov+default_base_moof+separate_moof \
  -f mp4 - | ./build/openmoq-publisher \
    --input - \
    --endpoint moqt://relay.example.com:443/moq \
    --namespace live/demo \
    --timeout 120

CTE LL-DASH HTTP Ingest

The DASH live path lets FFmpeg send CMAF/fMP4 over HTTP/1.1 chunked requests instead of piping fragmented MP4 through stdin. Start the publisher first:

./build/openmoq-publisher \
  --live-source dash \
  --dash-listen 0.0.0.0:8080 \
  --dash-path /ingest \
  --endpoint https://127.0.0.1:4433/moq \
  --transport webtransport \
  --namespace live \
  --draft 18 \
  --publish-catalog \
  --forward 1 \
  --insecure

Then push a live DASH feed to the ingest prefix:

ffmpeg -re \
  -f lavfi -i "testsrc2=size=1280x720:rate=25" \
  -f lavfi -i "anullsrc=r=48000:cl=stereo" \
  -filter_complex "[0:v]split=2[v1][v2];[v1]scale=1280:720[v720];[v2]scale=640:360[v360]" \
  -map "[v720]" -c:v:0 libx264 -b:v:0 1500k -g 50 -keyint_min 50 -sc_threshold 0 \
  -map "[v360]" -c:v:1 libx264 -b:v:1 500k -g 50 -keyint_min 50 -sc_threshold 0 \
  -map 1:a -c:a aac -b:a 128k \
  -f dash -seg_duration 2 -use_template 1 -use_timeline 0 \
  -init_seg_name 'video$RepresentationID$' \
  -media_seg_name 'video$RepresentationID$' \
  -adaptation_sets "id=0,streams=v id=1,streams=a" \
  -multiple_requests 1 -streaming 1 -remove_at_exit 0 \
  -window_size 20 -extra_window_size 20 \
  http://127.0.0.1:8080/ingest/

With this naming pattern, FFmpeg sends representation requests under the ingest prefix, such as /ingest/video0, /ingest/video1, and /ingest/video2. The publisher treats each representation path independently, discovers tracks from init segments sent on those paths, and emits catalog plus media objects for subscribers.

For relay smoke testing, use --forward 1 so objects are forwarded immediately. Use --forward 0 when the relay should wait for subscriber interest before media delivery. In that mode, connection_id= confirms transport and MOQT setup only; it does not mean the relay accepted the namespace or forwarded a subscription. Draft-16 subscriptions arrive on the control stream, while draft-17/18 subscriptions arrive on bidirectional request streams and are acknowledged with SUBSCRIBE_OK on the same stream. Use --timeout to bound the initial wait and OPENMOQ_PICOQUIC_TRACE=1 to inspect relay control traffic.