Skip to content

Video Transcoding

AVC Encoders

The encoder used by FFmpeg can be configured with PHOTOPRISM_FFMPEG_ENCODER in your docker-compose.yml config file:

Encoder Value
Software H.264 software
Apple Video Toolbox apple
Intel Quick Sync intel
NVIDIA H.264 nvidia
Raspberry Pi / Video4Linux raspberry
Video Acceleration API vaapi

It defaults to software if no value is set or hardware transcoding fails. Please refer to the FFmpeg documentation for a full list of encoders and their implementation status. We welcome contributions to support additional encoders.

For video transcoding to work, FFmpeg must be enabled and installed. When using our Docker images, it is already pre-installed. In addition, the service must have permission to use the related video devices. This depends on your hardware and operating system, so we can only give you examples that may need to be changed to work for you.

Size Limit

The PHOTOPRISM_FFMPEG_SIZE config option allows to limit the resolution of transcoded videos. It accepts the following standard sizes, while other values are automatically adjusted to the next supported size:

Size Usage
720 SD TV, Mobile
1280 HD TV, SXGA
1920 Full HD
2048 DCI 2K, Tablets
2560 Quad HD, Notebooks
3840 4K Ultra HD
4096 DCI 4K, Retina 4K
7680 8K Ultra HD 2

When transcoding videos, the original aspect ratio is maintained and smaller videos will not be upscaled.

Note that MPEG-4 AVC videos are not re-encoded if they exceed the configured resolution limit.

Bitrate Limit

You can limit the bitrate of the AVC encoder with the config option PHOTOPRISM_FFMPEG_BITRATE. Keep in mind that this is a "soft limit", so the actual bitrate varies and depends on the encoder used as well as the specific FFmpeg parameters, which in turn depend on the encoder. It may also depend on the operating system and the GPU drivers.

If the bitrate is significantly exceeded in your environment and you want improvements to be implemented, we recommend that you take a look at the FFmpeg documentation and the parameters in our source code so you can tell us which parameters should be changed to make it work for you.

Note that MPEG-4 AVC videos are not re-encoded if they exceed the configured bitrate limit. To reduce the size of AVC videos, you can manually replace the original files with a smaller version or wait for a future release that offers this functionality.

Already transcoded video files are not automatically re-transcoded when the limit is changed. To do this, you must manually remove the *.avc files in the sidecar storage folder and run the photoprism convert command in a terminal.

Software Transcoding

Unless you have a lot of high-resolution videos in your library, we recommend keeping the default settings to use the standard software codec for video transcoding. It has a high quality and does not require any special permissions or additional drivers.

Since FFmpeg 7 has significant performance optimizations compared to version 6.1.1 that ships with Ubuntu 24.04, users of our Docker image can choose to install the latest FFmpeg build available at by adding PHOTOPRISM_INIT: "ffmpeg" to the environment section of their compose.yaml or docker-compose.yml file:

      PHOTOPRISM_INIT: "ffmpeg"

Note that this version cannot be used with hardware transcoding and that it may support a different set of file formats.

GPU Drivers

Depending on your hardware, it may be necessary to install additional packages for FFmpeg to use the AVC encoding device.

One way to do this automatically is to set PHOTOPRISM_INIT to "gpu tensorflow" when using our Docker images. Note that this is experimental and not required for most encoders.

See the related installation script on GitHub for details. We welcome contributions to support additional devices or update package names if needed.

Most users can either skip PHOTOPRISM_INIT completely or just use PHOTOPRISM_INIT: "tensorflow" to install a special version of TensorFlow that improves indexing performance if the server CPU supports AVX, which is independent of video transcoding and the type of GPU.

Intel Quick Sync

To enable Intel Quick Sync hardware video transcoding, add the intel target to PHOTOPRISM_INIT, choose the intel encoder, and share the /dev/dri devices with the photoprism service:

      PHOTOPRISM_INIT: "intel"
      - "/dev/dri:/dev/dri"
      - ...

In addition, you can choose to run the photoprism service as a non-root user by setting either the user service property or the PHOTOPRISM_UID and PHOTOPRISM_GID environment variables in your docker-compose.yml file:

Environment Default Description
PHOTOPRISM_UID 0 run as a non-root user after initialization (supported: 0, 33, 50-99, 500-600, 900-1250, and 2000-2100)
PHOTOPRISM_GID 0 run with a specific group id after initialization, can optionally be used together with PHOTOPRISM_UID (supported: 0, 33, 44, 50-99, 105, 109, 115, 116, 500-600, 900-1250, and 2000-2100)

Which user and group you choose should depend on the owner of the /dev/dri video device so that the service has permission to access it.

Finally, remember to update the file permissions and/or owner with the chmod and chown commands when you make changes to the UID or GID, and restart the services for your changes to take effect:

docker compose stop
docker compose up -d

Older Intel hardware may not support certain video codecs and resolutions. In this case, it is not possible to use hardware transcoding for these videos. We may later add a configuration option that allows you to downscale videos.

NVIDIA Container Toolkit

For hardware transcoding with an NVIDIA graphics card, the NVIDIA Container Toolkit must be installed on the host computer first. Instructions can be found in their installation guide.

Once the toolkit is installed, choose the nvidia encoder and add a deploy section to the photoprism service:

      PHOTOPRISM_INIT: "tensorflow"
      NVIDIA_DRIVER_CAPABILITIES: "compute,video,utility"
      - ...
            - driver: "nvidia"
              count: 1
              capabilities: [gpu]

Now restart the services for your changes to take effect:

docker compose stop
docker compose up -d

We also provide a ready-to-use docker-compose.yml example for your convenience. Note that older hardware may not support certain video codecs and resolutions.

Raspberry Pi

Experimental hardware-accelerated transcoding on a Raspberry Pi (and compatible devices) can be enabled by choosing the raspberry encoder:


The Docker container must also have access to one or more video devices. For the raspberry encoder, for example, you add:

 - "/dev/video11:/dev/video11"

Additional advanced configuration options are available to improve stability if needed:

PHOTOPRISM_FFMPEG_BUFFERS: "64" # FFmpeg capture buffers (default: 32)

Now restart the services for your changes to take effect:

docker compose stop
docker compose up -d

Some server configurations, especially Raspberry Pi's, may experience memory allocation issues when using hardware acceleration. Carefully monitor your server's logs and increase the available GPU and/or CMA memory allocations if necessary. Note that the Raspberry Pi hardware currently only supports video resolutions up to 2160p.

Other Hardware

If you want to use other hardware for transcoding, choose the appropriate AVC encoder and share the required devices with the photoprism service, as shown in the examples above. Then restart the services for the changes to take effect.

Which devices need to be shared and whether additional drivers are required depends on your specific hardware. For more information, see the FFmpeg documentation.


Enabling Trace Log Mode

A good way to troubleshoot configuration issues is to increase the log level. To enable trace log mode, set PHOTOPRISM_LOG_LEVEL to "trace" in the environment: section of the photoprism service (or use the --trace flag when running the photoprism command directly):


Then restart all services for your changes to take effect:

docker compose stop
docker compose up -d

Viewing Docker Service Logs

You can run this command to check the server logs for warnings and errors, including the last 100 messages (omit --tail=100 to see them all, and -f to output only the last logs without watching them):

docker compose logs -f --tail=100 

Learn more ›

If FFmpeg is disabled or not installed, videos cannot be indexed because still images cannot be created. You should also have ExifTool enabled to extract metadata such as duration, resolution, and codec. Note that your hardware may not support certain video codecs and resolutions. In this case, the software encoder is used automatically.

Our examples use the new docker compose command by default. If your server does not yet support it, you can still use docker-compose or alternatively podman-compose on Red Hat-compatible distributions.

Help improve these docs! You can contribute by clicking to send a pull request with your changes.