3pp-mirror/neko
1
0
Fork 0
mirror of https://github.com/m1k1o/neko.git synced 2025-04-28 18:06:20 +02:00
Code Issues Projects Releases Packages Wiki Activity
neko/webpage/docs/configuration/capture.md
Miroslav Šedivý 81c259cdc9 docs: add nvidis gpu examples. #502
2025-04-05 21:45:19 +02:00

21 KiB
Raw Blame History

description
Configuration related to Gstreamer capture in Neko.

import { Def, Opt } from '@site/src/components/Anchor'; import { ConfigurationTab } from '@site/src/components/Configuration'; import configOptions from './help.json';

Audio & Video Capture

This guide will show you how to configure the audio and video capture settings in neko.

Neko uses Gstreamer to capture and encode audio and video in the following scenarios:

  • WebRTC clients use the Video and Audio pipelines to receive the audio and video streams from the server.
  • The Broadcast feature allows you to broadcast the audio and video to a third-party service using RTMP.
  • The WebRTC Fallback mechanism allows you to capture the display in the form of JPEG images and serve them over HTTP using Screencast.
  • Clients can share their Webcam and Microphone with the server using WebRTC.

WebRTC Video

Neko allows you to capture the display and encode it in real-time using Gstreamer. The encoded video is then sent to the client using WebRTC. This allows you to share the display with the client in real-time.

There can exist multiple video pipelines in neko that are referenced by their unique pipeline id. Each video pipeline can have its own configuration settings and clients can either choose which pipeline they want to use or let neko choose the best pipeline for them.

:::info Limitation All video pipelines must use the same video codec (defined in the setting). :::

The Gstreamer pipeline is started when the first client requests the video stream and is stopped after the last client disconnects.

<ConfigurationTab options={configOptions} filter={[ "capture.video.display", "capture.video.codec", "capture.video.ids", "capture.video.pipeline", "capture.video.pipelines", ]} comments={false} />

  • is the name of the X display that you want to capture. If not specified, the environment variable DISPLAY will be used.
  • available codecs are vp8, vp9, av1, h264. Supported video codecs are dependent on the WebRTC implementation used by the client, vp8 and h264 are supported by all WebRTC implementations.
  • is a list of pipeline ids that are defined in the section. The first pipeline in the list will be the default pipeline.
  • is a shorthand for defining Gstreamer pipeline description for a single pipeline. This is option is ignored if is defined.
  • is a dictionary of pipeline configurations. Each pipeline configuration is defined by a unique pipeline id. They can be defined in two ways: either by building the pipeline dynamically using Expression-Driven Configuration or by defining the pipeline using a Gstreamer Pipeline Description.

Expression-Driven Configuration

Expression allows you to build the pipeline dynamically based on the current resolution and framerate of the display. Expressions are evaluated using the gval library. Available variables are , , and of the display at the time of capture.

capture:
  video:
    ...
    pipelines:
      <pipeline_id>:
        width: "<expression>"
        height: "<expression>"
        fps: "<expression>"
        gst_prefix: "<gst_pipeline>"
        gst_encoder: "<gst_encoder_name>"
        gst_params:
          <param_name>: "<expression>"
        gst_suffix: "<gst_pipeline>"
        show_pointer: true
  • , , and are the expressions that are evaluated to get the stream resolution and framerate. They can be different from the display resolution and framerate if downscaling or upscaling is desired.
  • and allow you to add custom Gstreamer elements before and after the encoder. Both parameters need to start with ! and then be followed by the Gstreamer elements.
  • is the name of the Gstreamer encoder element, such as vp8enc or x264enc.
  • are the parameters that are passed to the encoder element specified in .
  • is a boolean value that determines whether the mouse pointer should be captured or not.
Example pipeline configuration

import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem';

```yaml title="config.yaml"
capture:
  video:
    codec: vp8
    # HQ is the default pipeline
    ids: [ hq, lq ]
    pipelines:
      hq:
        fps: 25
        gst_encoder: vp8enc
        gst_params:
          target-bitrate: round(3072 * 650)
          cpu-used: 4
          end-usage: cbr
          threads: 4
          deadline: 1
          undershoot: 95
          buffer-size: (3072 * 4)
          buffer-initial-size: (3072 * 2)
          buffer-optimal-size: (3072 * 3)
          keyframe-max-dist: 25
          min-quantizer: 4
          max-quantizer: 20
      lq:
        fps: 25
        gst_encoder: vp8enc
        gst_params:
          target-bitrate: round(1024 * 650)
          cpu-used: 4
          end-usage: cbr
          threads: 4
          deadline: 1
          undershoot: 95
          buffer-size: (1024 * 4)
          buffer-initial-size: (1024 * 2)
          buffer-optimal-size: (1024 * 3)
          keyframe-max-dist: 25
          min-quantizer: 4
          max-quantizer: 20
```

```yaml title="config.yaml"
capture:
  video:
    codec: h264
    ids: [ main ]
    pipelines:
      main:
        width: (width / 3) * 2
        height: (height / 3) * 2
        fps: 20
        gst_prefix: "! video/x-raw,format=I420"
        gst_encoder: "x264enc"
        gst_params:
          threads: 4
          bitrate: 4096
          key-int-max: 15
          byte-stream: true
          tune: zerolatency
          speed-preset: veryfast 
        gst_suffix: "! video/x-h264,stream-format=byte-stream"
```

Gstreamer Pipeline Description

If you want to define the pipeline using a Gstreamer pipeline description, you can do so by setting the parameter.

capture:
  video:
    ...
    pipelines:
      <pipeline_id>:
        gst_pipeline: "<gstreamer_pipeline>"

Since now you have to define the whole pipeline, you need to specify the src element to get the video frames and the sink element to send the encoded video frames to neko. In your pipeline, you can use {display} as a placeholder for the display name that will be replaced by the actual display name at runtime. You need to set the name property of the sink element to appsink so that neko can capture the video frames.

Your typical pipeline string would look like this:

ximagesrc display-name={display} show-pointer=true use-damage=false ! <your_elements> ! appsink name=appsink

See documentation for ximagesrc and appsink for more information.

Example pipeline configuration 
```yaml title="config.yaml"
capture:
  video:
    codec: vp8
    ids: [ hq, lq ]
    pipelines:
      hq:
        gst_pipeline: |
          ximagesrc display-name={display} show-pointer=true use-damage=false
          ! videoconvert ! queue
          ! vp8enc
            name=encoder
            target-bitrate=3072000
            cpu-used=4
            end-usage=cbr
            threads=4
            deadline=1
            undershoot=95
            buffer-size=12288
            buffer-initial-size=6144
            buffer-optimal-size=9216
            keyframe-max-dist=25
            min-quantizer=4
            max-quantizer=20
          ! appsink name=appsink
      lq:
        gst_pipeline: |
          ximagesrc display-name={display} show-pointer=true use-damage=false
          ! videoconvert ! queue
          ! vp8enc
            name=encoder
            target-bitrate=1024000
            cpu-used=4
            end-usage=cbr
            threads=4
            deadline=1
            undershoot=95
            buffer-size=4096
            buffer-initial-size=2048
            buffer-optimal-size=3072
            keyframe-max-dist=25
            min-quantizer=4
            max-quantizer=20
          ! appsink name=appsink
```

```yaml title="config.yaml"
capture:
  video:
    codec: h264
    ids: [ main ]
    pipelines:
      main:
        gst_pipeline: |
          ximagesrc display-name={display} show-pointer=true use-damage=false
          ! videoconvert ! queue
          ! x264enc
            name=encoder
            threads=4
            bitrate=4096
            key-int-max=15
            byte-stream=true
            tune=zerolatency
            speed-preset=veryfast
          ! video/x-h264,stream-format=byte-stream
          ! appsink name=appsink
```

```yaml title="config.yaml"
capture:
  video:
    codec: h264
    ids: [ main ]
    pipelines:
      main:
        gst_pipeline: |
          ximagesrc display-name={display} show-pointer=true use-damage=false
          ! videoconvert ! queue
          ! video/x-raw,format=NV12
          ! nvh264enc
            name=encoder
            preset=2
            gop-size=25
            spatial-aq=true
            temporal-aq=true
            bitrate=4096
            vbv-buffer-size=4096
            rc-mode=6
          ! h264parse config-interval=-1
          ! video/x-h264,stream-format=byte-stream
          ! appsink name=appsink
```

This configuration requires [Nvidia GPU](https://developer.nvidia.com/cuda-gpus) with [NVENC](https://developer.nvidia.com/nvidia-video-codec-sdk) support.

WebRTC Audio

Only one audio pipeline can be defined in neko. The audio pipeline is used to capture and encode audio, similar to the video pipeline. The encoded audio is then sent to the client using WebRTC.

The Gstreamer pipeline is started when the first client requests the video stream and is stopped after the last client disconnects.

<ConfigurationTab options={configOptions} filter={[ "capture.audio.device", "capture.audio.codec", "capture.audio.pipeline", ]} comments={false} />

  • is the name of the pulseaudio device that you want to capture. If not specified, the default audio device will be used.
  • available codecs are opus, g722, pcmu, pcma. Supported audio codecs are dependent on the WebRTC implementation used by the client, opus is supported by all WebRTC implementations.
  • is the Gstreamer pipeline description that is used to capture and encode audio. You can use {device} as a placeholder for the audio device name that will be replaced by the actual device name at runtime.
Example pipeline configuration 
capture:
  audio:
    codec: opus
    pipeline: |
      pulsesrc device={device}
      ! audioconvert
      ! opusenc
        bitrate=320000
      ! appsink name=appsink

Broadcast

Neko allows you to broadcast out-of-the-box the display and audio capture to a third-party service. This can be used to broadcast the display and audio to a streaming service like Twitch or YouTube, or to a custom RTMP server like OBS, Nginx RTMP module, or MediaMTX.

The Gstreamer pipeline is started when the broadcast is started and is stopped when the broadcast is stopped regardless of the clients connected.

<ConfigurationTab options={configOptions} filter={[ "capture.broadcast.audio_bitrate", "capture.broadcast.video_bitrate", "capture.broadcast.preset", "capture.broadcast.pipeline", "capture.broadcast.url", "capture.broadcast.autostart", ]} comments={false} />

The default encoder uses h264 for video and aac for audio, muxed in the flv container and sent over the rtmp protocol. You can change the encoder settings by setting a custom Gstreamer pipeline description in the parameter.

  • and are the bitrate settings for the default audio and video encoders expressed in kilobits per second.
  • is the encoding speed preset for the default video encoder. See available presets here.
  • when set, encoder settings above are ignored and the custom Gstreamer pipeline description is used. In the pipeline, you can use {display}, {device} and {url} as placeholders for the X display name, pulseaudio audio device name, and broadcast URL respectively.
  • is the URL of the RTMP server where the broadcast will be sent e.g. rtmp://<server>/<application>/<stream_key>. This can be set later using the API if the URL is not known at the time of configuration or is expected to change.
  • is a boolean value that determines whether the broadcast should start automatically when neko starts, works only if the URL is set.
Example pipeline configuration 
```yaml title="config.yaml"
capture:
  broadcast:
    pipeline: |
      flvmux name=mux
        ! rtmpsink location={url}
      pulsesrc device={device}
        ! audio/x-raw,channels=2
        ! audioconvert
        ! voaacenc
        ! mux.
      ximagesrc display-name={display} show-pointer=false use-damage=false
        ! video/x-raw,framerate=28/1
        ! videoconvert
        ! queue
        ! x264enc bframes=0 key-int-max=0 byte-stream=true tune=zerolatency speed-preset=veryfast
        ! mux.
```

```yaml title="config.yaml"
capture:
  broadcast:
    pipeline: |
      flvmux name=mux
        ! rtmpsink location={url}
      pulsesrc device={device}
        ! audio/x-raw,channels=2
        ! audioconvert
        ! voaacenc
        ! mux.
      ximagesrc display-name={display} show-pointer=false use-damage=false
        ! video/x-raw,framerate=30/1
        ! videoconvert
        ! queue
        ! video/x-raw,format=NV12
        ! nvh264enc name=encoder preset=low-latency-hq gop-size=25 spatial-aq=true temporal-aq=true bitrate=2800 vbv-buffer-size=2800 rc-mode=6
        ! h264parse config-interval=-1
        ! video/x-h264,stream-format=byte-stream,profile=high
        ! h264parse
        ! mux.
```

This configuration requires [Nvidia GPU](https://developer.nvidia.com/cuda-gpus) with [NVENC](https://developer.nvidia.com/nvidia-video-codec-sdk) support and [Nvidia docker image](/docs/v3/installation/docker-images#nvidia) of neko.

Screencast

As a fallback mechanism, neko can capture the display in the form of JPEG images and the client can request these images over HTTP. This is useful when the client does not support WebRTC or when the client is not able to establish a WebRTC connection, or there is a temporary issue with the WebRTC connection and the client should not miss the content being shared.

:::note This is a fallback mechanism and should not be used as a primary video stream because of the high latency, low quality, and high bandwidth requirements. :::

The Gstreamer pipeline is started in the background when the first client requests the screencast and is stopped after a period of inactivity.

<ConfigurationTab options={configOptions} filter={[ "capture.screencast.enabled", "capture.screencast.rate", "capture.screencast.quality", "capture.screencast.pipeline", ]} comments={false} />

  • is a boolean value that determines whether the screencast is enabled or not.
  • is the framerate of the screencast. It is expressed as a fraction of frames per second, for example, 10/1 means 10 frames per second.
  • is the quality of the JPEG images. It is expressed as a percentage, for example, 60 means 60% quality.
  • when set, the default pipeline settings above are ignored and the custom Gstreamer pipeline description is used. In the pipeline, you can use {display} as a placeholder for the X display name.
Example pipeline configuration 
capture:
  screencast:
    enabled: true
    pipeline: |
      ximagesrc display-name={display} show-pointer=true use-damage=false
        ! video/x-raw,framerate=10/1
        ! videoconvert
        ! queue
        ! jpegenc quality=60
        ! appsink name=appsink

Webcam

:::danger This feature is experimental and may not work on all platforms. :::

Neko allows you to capture the webcam on the client machine and send it to the server using WebRTC. This can be used to share the webcam feed with the server.

The Gstreamer pipeline is started when the client shares their webcam and is stopped when the client stops sharing the webcam. Maximum one webcam pipeline can be active at a time.

<ConfigurationTab options={configOptions} filter={[ "capture.webcam.enabled", "capture.webcam.device", "capture.webcam.width", "capture.webcam.height", ]} comments={false} />

  • is a boolean value that determines whether the webcam capture is enabled or not.
  • is the name of the video4linux device that will be used as a virtual webcam.
  • and are the resolution of the virtual webcam feed.

In order to use the webcam feature, the server must have the v4l2loopback kernel module installed and loaded. The module can be loaded using the following command:

# Install the required packages (Debian/Ubuntu)
sudo apt install v4l2loopback-dkms v4l2loopback-utils linux-headers-`uname -r` linux-modules-extra-`uname -r`
# Load the module with exclusive_caps=1 to allow multiple applications to access the virtual webcam
sudo modprobe v4l2loopback exclusive_caps=1

This is needed even if neko is running inside a Docker container. In that case, the v4l2loopback module must be loaded on the host machine and the device must be mounted inside the container.

services:
  neko:
    ...
    # highlight-start
    devices:
      - /dev/video0:/dev/video0
    # highlight-end
    ...

Microphone

Neko allows you to capture the microphone on the client machine and send it to the server using WebRTC. This can be used to share the microphone feed with the server.

The Gstreamer pipeline is started when the client shares their microphone and is stopped when the client stops sharing the microphone. Maximum one microphone pipeline can be active at a time.

<ConfigurationTab options={configOptions} filter={[ "capture.microphone.enabled", "capture.microphone.device", ]} comments={false} />

  • is a boolean value that determines whether the microphone capture is enabled or not.
  • is the name of the pulseaudio device that will be used as a virtual microphone.