Installation

Two components: the Jellyfin plugin (DLL loaded by the server) and the Docker AI service (containerised ONNX runtime). Both are required.

1 · Requirements

2 · Install the plugin (repository)

Jellyfin's plugin manager can pull from any HTTPS JSON manifest. This is the recommended method because updates install automatically.

1. Open Jellyfin Dashboard → Plugins → Repositories → +.
2. Add a repository with:
   • Name: AI Upscaler
   • URL:
     https://raw.githubusercontent.com/Kuschel-code/JellyfinUpscalerPlugin/main/repository-jellyfin.json
3. Click Save, then go to Plugins → Catalog. The AI Upscaler Plugin entry appears under Video Enhancement.
4. Click Install, then restart Jellyfin when prompted.
5. Confirm under Plugins → My Plugins the version reads 1.6.1.16.

3 · Install the plugin (manual ZIP)

Use this if your Jellyfin host can't reach GitHub, or you want to pin a specific version.

1. Download JellyfinUpscalerPlugin-v1.6.1.16.zip from GitHub Releases.
2. Extract into Jellyfin's plugin folder: <config>/plugins/AI Upscaler Plugin_1.6.1.4/. The folder name is historical — don't rename it.
3. chown the extracted files to Jellyfin's UID/GID (typically 568:568 on TrueNAS SCALE, 105:108 on LinuxServer.io images).
4. Restart Jellyfin. The plugin will load the DLL + shipped transitive NuGet dependencies (CliWrap.dll, FFMpegCore.dll, Instances.dll, SixLabors.ImageSharp.dll).

4 · Install the Docker AI service

Four image variants are published on Docker Hub — pick the one that matches your GPU.

# NVIDIA CUDA (Turing+, 6GB+ VRAM recommended)
docker run -d --name jellyfin-ai-upscaler \
  --gpus all -p 5000:5000 \
  -v upscaler-models:/app/models \
  -e API_TOKEN=change-me-please \
  kuscheltier/jellyfin-ai-upscaler:latest-cuda

# Intel OpenVINO (Arc, iGPU, N100)
docker run -d --name jellyfin-ai-upscaler \
  --device /dev/dri -p 5000:5000 \
  -v upscaler-models:/app/models \
  -e API_TOKEN=change-me-please \
  kuscheltier/jellyfin-ai-upscaler:latest-openvino

# AMD ROCm (RDNA2+)
docker run -d --name jellyfin-ai-upscaler \
  --device=/dev/kfd --device=/dev/dri \
  --group-add video -p 5000:5000 \
  -v upscaler-models:/app/models \
  -e API_TOKEN=change-me-please \
  kuscheltier/jellyfin-ai-upscaler:latest-rocm

# CPU-only fallback (works everywhere, slow)
docker run -d --name jellyfin-ai-upscaler \
  -p 5000:5000 \
  -v upscaler-models:/app/models \
  -e API_TOKEN=change-me-please \
  kuscheltier/jellyfin-ai-upscaler:latest-cpu

5 · Wire plugin → AI service

1. Open the plugin config: Dashboard → Plugins → AI Upscaler Plugin → Settings.
2. Under AI Service:
   • URL: http://<docker-host>:5000
   • API Token: the API_TOKEN value from the docker run above (leave blank if you omitted it)
3. Click Test Connection. You should see a green Online badge and the real auth posture (token-configured vs auth-disabled).
4. Save. The plugin will immediately call /health, /models, /hardware, and /features to populate the dashboard.

6 · Download models

Models are not shipped in the Docker image — they're pulled on demand to keep the base image small.

1. Open the plugin config → AI Models tab.
2. Pick a starter set (the defaults suit most libraries): realesrgan-x4plus (general), animesr-v2-x4 (anime), fsrcnn-x2 (fast, low-VRAM).
3. Click Download — the service pulls ONNX weights to /app/models/ (persisted on your mount volume).
4. Once downloaded, click Load to warm the model in GPU memory.

7 · Verify

1. Open any video in Jellyfin, click the AI Upscaler button that appears next to the subtitle selector.
2. You should see the tabbed quick-menu (Models / Filters / Realtime) with model icons reflecting their state (downloaded, loaded, busy).
3. Pick a model → Apply → watch the LIVE indicator pulse green while the filter is active.