mythos

Docker images, title search, and a new player

Four pieces of work since Phase 3 landed: a published Docker image, a title-search box across movies and series, a media-chrome player rebuild, and the Tonemapx CPU pipeline.

· status

A handful of things landed in main since the Phase 3 + HDR→SDR post — none of them open a new phase, but together they take a real chunk out of the “early development” rough edges.

Docker images on Docker Hub

The fastest way to try Mythos is now a docker pull. Multi-arch images (linux/amd64 and linux/arm64) get rebuilt and pushed on every commit to main:

docker run -d --name mythos \
  -p 8080:8080 \
  -v mythos-data:/data \
  -v /path/to/media:/media:ro \
  -e MYTHOS_TMDB_API_KEY=... \
  darkspar/mythos-server:edge

A few opinions baked into the image:

  • jellyfin-ffmpeg7 bundled. That gets you the tonemap_vaapi / tonemap_opencl / tonemap_cuda HW filters plus tonemapx (more on that below) without having to hunt them down distro-by-distro. The image already points MYTHOS_FFMPEG_BIN / MYTHOS_FFPROBE_BIN at it.
  • PID 1 is tini. Signals reach mythos-server cleanly; reaping is handled.
  • /data is the single writable volume. SQLite DB, JWT secret, poster cache, transcode segments, sidecar subtitles — everything Mythos writes lands there. Bind-mount your media read-only.

Tags follow Docker conventions: :edge is rolling, :sha-<short> is immutable per-commit (pin this for production-ish setups so an unattended docker pull doesn’t surprise you), and semver tags will appear when the first v* git tag is pushed. There’s intentionally no :latest yet — by Docker convention that means “newest stable release,” and Mythos hasn’t cut one.

Title search across the library

A single search box. One endpoint, GET /api/search?q=…, returns a flat ranked list across movies and series. The implementation is deliberately humble — case-insensitive LIKE against sort_title via mythos_db::SearchRepo — and will graduate to SQLite FTS5 once libraries get big enough to chug. For now it’s instant on the libraries people actually have.

The UI is the part I’m happier with. Results render inline as you type; ↑/↓ walks the list, Enter opens, and Esc dismisses. There’s no separate “search page” — the results panel sits under the box on whatever page you’re on, and the keyboard never leaves the input.

A new player, built on media-chrome

The old player was the native <video controls> element with a couple of overlays. It worked, but customising the chrome ranged from hard-to-impossible across browsers, and the more features Mythos grew (continue-watching cards, auto-play-next countdowns, an in-player CC picker) the worse the seams showed.

Player.svelte is now rebuilt on top of media-chrome — a set of Web Components that wrap a plain <video> element with named slots for every control. The video element underneath is still a real HTMLVideoElement; media-chrome is purely a chrome layer.

What that unlocked:

  • YouTube/Netflix-flavoured chrome. A two-bar layout with all controls visible; the volume slider is a hover-reveal so it doesn’t clutter mobile; the close button lives on the player’s top-right corner.
  • An actual CC menu. A subtitle picker as a fixed-position panel, not a hard-coded <track> dropdown. Toggling it no longer pauses the video.
  • Click / dblclick gestures on the video frame. Single click toggles play; double click toggles fullscreen. media-chrome’s built-in gesture receiver is disabled so they don’t double-fire with our own.
  • Hover-reveal play buttons on tiles and detail pages. A poster art hover now shows a Play affordance; movie + episode tiles have a watch-progress bar baked in.

Tonemapx — SIMD CPU tonemap

The newest pipeline option in the HDR→SDR settings is Tonemapx, jellyfin-ffmpeg’s SIMD-optimised replacement for the stock CPU tonemap filter. On the CPU path it’s dramatically faster than software — fast enough that a CPU-only box can keep up with HDR sources where it previously couldn’t.

The catch: tonemapx is jellyfin-ffmpeg-only on most distros. Mythos probes for it at startup, so picking it from the admin UI on a stock distro ffmpeg silently falls back to software rather than breaking playback. The Docker image already ships jellyfin-ffmpeg and points the ffmpeg binary env vars at it; on bare-metal installs set MYTHOS_FFMPEG_BIN=/usr/lib/jellyfin-ffmpeg/ffmpeg (and the matching ffprobe) to light it up.

The pipeline list is now: software, tonemapx, vaapi, opencl, cuda. NVENC stays on the GPU end-to-end via NVDEC + scale_cuda + tonemap_cuda; QSV / VideoToolbox boxes get the SIMD CPU kernel inline where there’s no GPU filter surface to download to.

What’s next

Still Phase 3: music, photos, books — separate sub-phases. After that, Phase 6 lights up the Jellyfin-API compatibility shim so existing clients like Findroid and Swiftfin work against Mythos without changes.

Source on GitLab .