Probe, decode, encode and make MP4s from frames with pcod

pcod is a small command line over the library. probe shows what is inside an MP4, decode decodes its whole video track and reports the speed, encode decodes, encodes and muxes a new MP4, copying the audio, and frames encodes a folder of PNG frames and a WAV into an MP4. Every option is in the command reference.

The outputs below are from a 10-second 1080p60 H.264 clip cut from NASA’s public-domain “A Message of Resilience from the International Space Station”, on a 16-core CPU and an AMD card through VA-API. Your speeds will differ.

Probe

pcod probe clip.mp4

container  MP4
duration   10.0 s
video      h264 High 1920x1080 59.94 fps
audio      aac 2 ch 48000 Hz

Decode

--route picks the GPU (the default) or the CPU; --threads caps the CPU threads.

pcod decode clip.mp4 --route cpu
pcod decode clip.mp4 --route gpu

route   CPU (16 threads)
decoded 600 frames in 0.88 s: 686 fps
route   GPU (VA-API, radeonsi)
decoded 600 frames in 0.48 s: 1260 fps

Without a GPU decoder for the file’s codec, --route gpu decodes on the CPU and says why on a note line.

Encode

--codec is h264 (the default) or hevc, --crf sets the quality (23 by default; lower is better and larger), and --route picks the GPU or the CPU for both the decode and the encode. On the GPU, frames go from the decoder to the encoder without leaving the card.

pcod encode clip.mp4 out.mp4 --codec hevc --route gpu

decode  GPU (VA-API, radeonsi)
encode  HEVC GPU (VA-API, radeonsi)
encoded 600 frames in 1.14 s: 526 fps
wrote   out.mp4 (2.7 MB)

HEVC encodes on the GPU only. Asked for it on the CPU, pcod encodes H.264 and says so:

pcod encode clip.mp4 out.mp4 --codec hevc --route cpu

decode  CPU (16 threads)
note    HEVC encodes on the GPU only; encoding H.264 on the CPU instead
encode  H.264 CPU (16 threads)
encoded 600 frames in 6.29 s: 95 fps
wrote   out.mp4 (4.2 MB)

Frames to an MP4

frames encodes a folder of PNG frames, such as a game’s rendered trailer, and an optional WAV into an MP4: H.264 (or HEVC with --codec hevc), 8-bit 4:2:0 BT.709, AAC at 48 kHz stereo, with the index at the front (faststart). The frames are the folder’s .png files in name order (000001.png, 000002.png, …), all one even size. --fps takes an integer, a ratio such as 30000/1001 or a decimal. The PNGs are decoded on every core ahead of the encoder. The encode goes to the GPU first and falls back to H.264 on the CPU, saying why on a note line.

pcod frames frames trailer.mp4 --fps 60 --audio trailer.wav

frames  600 PNG 1920x1080 at 60 fps (10.00 s), 16 decode threads
audio   trailer.wav: 10.00 s, AAC 48 kHz stereo
encode  H.264 GPU
encoded 600 frames in 1.07 s: 560 fps
wrote   trailer.mp4 (7.5 MB)

These are the same 10 seconds of the NASA clip, written out as 1080p PNG frames. The WAV can be 8, 16, 24 or 32-bit PCM or float, at any rate, mono, stereo or more channels. pcod converts it to 48 kHz stereo and cuts or pads it with silence to the length of the video. Without --audio, the MP4 carries a silent stereo track. A missing frame number, or a frame of another size, stops the encode:

pcod: gap: the frames jump from 000001.png to 000003.png

When something fails

A file pcod can’t read, or a failing route, ends with one line on stderr and exit code 1:

pcod: missing.mp4: No such file or directory

A command line it can’t parse prints clap’s error and usage on stderr with exit code 2; pcod --help and pcod help <command> print the full help.

pcod’s output is plain text for people; it has no JSON mode. For anything beyond these four commands, use the library.