Recherche avancée

Sur d’autres sites (8668)

• How do I bundle ffmpeg with my PyInstaller —onefile executable correctly ?

  8 septembre, par user31448173

  I’ve been struggling for a while to package my Python script with PyInstaller into a standalone executable, but the main blocker seems to be getting ffmpeg and yt_dlp bundled and accessible in the final build.

  


  My Goal

  


  I want a true standalone executable created with PyInstaller (—onefile) that includes ffmpeg so that it works out-of-the-box on any machine, without requiring users to install ffmpeg or set up environment variables.

  


  Question

  


  How can I properly bundle ffmpeg into my PyInstaller build ?

  


  Has anyone successfully bundled ffmpeg with —onefile and can share the working approach ? Or point me in the right direction ?

  


  Additional Info

  


  OS : Windows
Python version : Python 3.13.0
PyInstaller version : 6.15.0
ffmpeg version : 2025-09-01-git-3ea6c2fe25-full_build-www.gyan.dev (...)
built with gcc 15.2.0

  


  i tried Running :

  


  pyinstaller —onefile discordmusicbot.py

  


  i've tried with hiddenimports, .spec file i got it to run but for some reason ffmpeg never seemed to work even packing it as .exe
datas.append(('ffmpeg.exe', '.'))

  


  Adding ffmpeg to PATH on my development machine, also doesn't work, but would also defeat the purpose of making a standalone executable.
I also know about the .spec placing it in the wrong directory, but moving it didnt help.
i also tried Meipass, didnt work either.

  


  Using PyInstaller hooks (—add-binary) :

  


  pyinstaller —onefile —add-binary "path/to/ffmpeg.exe ;." discordmusicbot.py

  


• ffmpeg : fisheye -> rectilinear conversion with remap_opencl : green tinted output [closed]

  4 septembre, par thinkfat

  I'm having a bit of a problem converting a fisheye video from an Insta360 camera to rectilinear. I know how to do it with the v360 filter but obviously it's slow, just barely realtime. I'm trying to set up a filter chain that uses remap_opencl and vaapi to keep everyhing in hw frames. I did succeed, technically, but the output video has a green tint, which hints that there is some pixel format related issue.

  


  The whole thing is on Linux, (openSUSE Tumbleweed if thats of any interest), a KabyLake platform with Intel HD 620 GPU. Using the NEO OpenCL driver from Intel (24.something, last version that still supported those GPUs). It implements va-opencl media sharing.

  


  This is the chain I came up with :

  


      [0:v]hwmap=derive_device=opencl[vid];
    [1:v]hwupload[xm];
    [2:v]hwupload[ym];
    [vid][xm][ym]remap_opencl[out];
    [out]hwmap=derive_device=vaapi:reverse=1[vout]


  


  The input into the chain are from a vaapi h.264 decoder and two precomputed maps in PGM format. All good here. The chain works and produces an output video that shows the mapping worked and that all the frames make it through the chain. It's fast, too, about 5x realtime. But the output video has a greenish tint which tells me that somewhere in the chain there is a pixel format related hickup. Obviously I want to avoid costly intermediate CPU involvement, so hwdownload,hwupload kind of defeats the purpose.

  


  This is the command line :

  


      ffmpeg -init_hw_device opencl=oc0:0.0 -filter_hw_device oc0&#xA;    -init_hw_device" vaapi=va:/dev/dri/renderD128&#xA;    -vaapi_device /dev/dri/renderD128&#xA;    -hwaccel vaapi&#xA;    -hwaccel_output_format vaapi&#xA;    -i input.mp4&#xA;    -i xmap.pgm&#xA;    -i ymap.pgm&#xA;    -filter_complex <filter chain="chain">&#xA;    -map [vout]&#xA;    -c:v h264_vaapi&#xA;    -c:a copy&#xA;    -y output.mp4&#xA;</filter>

  


  I'm looking for a way to do an explicit pixel format conversion before/after mapping the frames between the different hardware contexts. I tried it on the vaapi side by adding a "scale_vaapi=format=", but whatever I put there, it completely breaks the filter chain because of impossible conversions between formats.

  


  This is the ffmpeg version :

  


  ffmpeg version N-120955-g6ce02bcc3a Copyright (c) 2000-2025 the FFmpeg developers

  


  I had to compile it manually because nothing I had on the distro supported opencl and va-opencl media sharing.

  


• Preserving alpha when re-encoding VP8/VP9 WebM in FFmpeg without specifying input codec [closed]

  29 août, par mstyura

  I'm trying to understand why re-encoding a VP8 or VP9 WebM with alpha channel produces black pixels instead of preserving transparency.

  


  I simplified the problem to re-encoding VP8 with alpha to VP9 with alpha :

  


  ffmpeg -i video-vp8-with-alpha.webm -c:v libvpx-vp9 video-vp8-with-alpha-bad.webm


  


  The output loses the alpha channel. ffprobe shows :

  


  Input #0, matroska,webm, from 'video-vp8-with-alpha-bad.webm':
  Metadata:
    ENCODER         : Lavf62.1.103
  Duration: 00:00:03.07, start: 0.017000, bitrate: 152 kb/s
  Stream #0:0: Video: vp9 (Profile 0), yuv420p(tv, bt470bg/bt709/bt709, progressive), 720x1560, SAR 1:1 DAR 6:13, 30 fps, 30 tbr, 1k tbn, start 0.067000
    Metadata:
      ENCODER         : Lavc62.8.101 libvpx-vp9
      ALPHA_MODE      : 1
      DURATION        : 00:00:03.066000000
  Stream #0:1: Audio: opus, 48000 Hz, stereo, fltp, start 0.017000
    Metadata:
      ENCODER         : Lavc62.8.101 libopus
      DURATION        : 00:00:03.049000000


  


  However, if I explicitly force the input codec :

  


  ffmpeg -c:v libvpx -i video-vp8-with-alpha.webm -c:v libvpx-vp9 video-vp8-with-alpha-good.webm


  


  The alpha channel is preserved. ffprobe shows :

  


  Input #0, matroska,webm, from 'video-vp8-with-alpha-good.webm':
  Metadata:
    ENCODER         : Lavf62.1.103
  Duration: 00:00:03.07, start: 0.017000, bitrate: 400 kb/s
  Stream #0:0: Video: vp9 (Profile 0), yuv420p(tv, unknown/bt709/bt709, progressive), 720x1560, SAR 1:1 DAR 6:13, 30 fps, 30 tbr, 1k tbn, start 0.067000
    Metadata:
      ENCODER         : Lavc62.8.101 libvpx-vp9
      ALPHA_MODE      : 1
      DURATION        : 00:00:03.066000000
  Stream #0:1: Audio: opus, 48000 Hz, stereo, fltp, start 0.017000
    Metadata:
      ENCODER         : Lavc62.8.101 libopus
      DURATION        : 00:00:03.049000000


  


  I don't know in advance whether the input is VP8 or VP9, and I don't want to run ffprobe before ffmpeg.

  


  How can I make ffmpeg automatically preserve the alpha channel from the input file without explicitly specifying the decoder ?

  


  The ffmpeg from commit 0828a3b636 ;

  


  Build of ffmpeg download from BtbN/FFmpeg-Builds