mediahound 0.2.0

Turn photos of your movie & music collection into a sleek, searchable web catalog.

🎬🎵 MediaHound

Turn photos of your movie and music collection into a sleek, searchable web catalog.

Point MediaHound at a folder of cover photos — DVDs, VHS, Blu-ray, CDs, vinyl, cassettes — or import a CSV. It identifies each item, pulls in cover art, genres, cast/artist, studio/label, runtime/tracklist and ratings, writes a short enticing intro, estimates the used resale value, links where to watch (movies) or listen (music), and generates a polished static website you can search, filter by 🎬 Movies / 🎵 Music, sort, and curate — with a password-protected admin mode.

Movies are identified/enriched via TMDB / OMDb / Wikidata + JustWatch; music via MusicBrainz + Cover Art Archive (open, zero-key) with keyless Spotify / Apple Music / YouTube Music links.

▶ Live demo — explore a sample catalog in your browser (admin password: changeme).

ℹ️ The demo shows real movie posters and album covers (hotlinked from IMDb/OMDb and Cover Art Archive / Apple) so you can see what a finished catalog looks like — no cover images are stored in this repo. Extra gallery photos are generated placeholders standing in for your own photos. Your real catalog pulls art from TMDB / OMDb / Wikidata (movies) and MusicBrainz / Cover Art Archive (music), or falls back to the photos you take.

• Runs for anybody with zero API keys — open-source OCR + open data by default.
• Offline-first — never contacts the internet unless you explicitly ask (--online).
• Static output — deploy anywhere (Netlify, GitHub Pages, S3, Vercel) or just open the HTML file.
• No secrets in the repo — keys live in a gitignored .env; your catalog is generated output.

MIT-licensed. Your photos, keys and catalog never get committed to this tool's repo.

---

See it in 30 seconds (no photos, no keys)

git clone https://github.com/jchirayath/mediahound && cd mediahound
pip install -e .

mediahound init demo
mediahound build --config demo/config.toml --mock   # generates a 10-title sample catalog
cd demo && python3 -m http.server 8000              # open http://localhost:8000

That's the screenshot above. Click 🔒 Admin and sign in with changeme to try the read/write admin tools. Everything you see is generated by --mock — no internet, no API keys.

Once published to PyPI you can skip the clone and just pip install mediahound (then mediahound init …). Maintainers: see RELEASING.md for the one-time publish setup.

---

Cataloguing your own collection

pip install -e ".[ocr]"     # adds the default OCR identifier
# Install the Tesseract engine for OCR:
#   macOS:  brew install tesseract
#   Debian: sudo apt-get install tesseract-ocr

mediahound init mysite      # scaffolds mysite/ (RawImages/{video,audio}/, config.toml, web template)

# Sort your cover photos by media type:
cp ~/Pictures/dvd-covers/*.jpg   mysite/RawImages/video/    # 🎬 movies (DVD/VHS/Blu-ray/LaserDisc)
cp ~/Pictures/album-covers/*.jpg mysite/RawImages/audio/    # 🎵 music  (CD/vinyl/cassette)

mediahound build --config mysite/config.toml --online   # identify + enrich (see Providers below)
cd mysite && python3 -m http.server 8000

Raw-image folder convention

Photos are sorted into media-type subfolders so MediaHound knows how to identify each item:

Folder	Media type	Identified/enriched via
RawImages/video/	🎬 movies	TMDB / OMDb / Wikidata + JustWatch
RawImages/audio/	🎵 music	MusicBrainz + Cover Art Archive + listen links
RawImages/ (root)	defaults to movies	—

(movies/ and music/ are accepted aliases.) Add more photos anytime and re-run build — only the new ones are processed (state is tracked by content hash in data/manifest.json).

Or import from a CSV (no photos)

mediahound import catalog.csv --config mysite/config.toml          # add rows offline
mediahound import catalog.csv --config mysite/config.toml --online # …and fetch cover art + metadata
mediahound export --config mysite/config.toml -o backup.csv        # dump the whole catalog back to CSV

Columns (case-insensitive; extras ignored): media_type, title, artist, director, year, format, label, studio, genres, rating, barcode, cover_url, intro. Only title is required — any missing fields are left blank (or filled by --online); even a one-column list of titles works. media_type is inferred (music if an artist is given, else movie). See examples/sample-import.csv.

Prefer a UI? Under mediahound serve --admin the admin screen has an ⬆ Import list button — paste or upload the same CSV, optionally tick enrich online, and the titles are added and the site rebuilt in place.

---

Features

The catalog

• Search title / genre / cast / studio / intro, sort by title, year, recently-added, value or rating.
• Filters: format, genre, studio, streaming service, language, category, seen / unseen.
• Dense, aligned cards showing poster, title·year, ★rating · format · runtime · language, genres, director + cast, studio, where-to-watch, intro hook, and estimated resale value.
• Clickable everything: a genre, person, or studio filters the grid to matching titles.
• Adjustable density — viewers pick how many movies per row; responsive on web & mobile.

Photos

• Multi-photo galleries — flip through every photo of a title with ‹ › arrows.
• Click-to-zoom lightbox; set any photo as the default; rotate photos (baked in on rebuild).
• Auto-uprights sideways/landscape cover photos to portrait.

Where to watch & resale

• Where to watch — is it on Netflix / Amazon Prime / Hulu? A clickable ▶ badge + pills link straight to the title (via JustWatch, no key). A filter narrows to a specific service.
• Resale value — a heuristic estimate plus a live link to eBay sold/completed listings.

Two views

• Default view — public, read-only.
• Admin view — password-protected, read/write. Edit a title's name, year, format, studio & distributor; move a title between 🎬 Movies and 🎵 Music; mark seen; rotate / set-default / delete a photo; delete a title; and configure the library name, description, logo, which fields are shown, and default columns.

Editing & persisting your changes

Your edits are recorded as small corrections (keyed by title id). There are two ways to make them permanent so they survive every future mediahound build — pick one:

A. Live admin server (recommended — zero manual steps)

mediahound serve --admin          # serves the site at http://127.0.0.1:8765

Open the site, unlock admin, and edit. Every change is written straight into data/corrections.json (and seen-overrides.json) as you go — the badge shows “✓ Saved to disk.” Click ↻ Rebuild to re-bake the catalog and reload. Because the edit is already in data/, the next mediahound build (and any re-query) keeps it — edits never revert. The write API is localhost-only and refuses cross-origin requests; never expose it publicly.

B. Static export (for read-only/CDN hosting like Netlify or GitHub Pages)

When the site is served as plain files (no admin server), edits live in your browser. Click Export changes / Export seen — the download is merged with the site's existing data/corrections.json so nothing already saved is lost — then drop the file into data/ and run mediahound build.

Either way the source of truth is data/corrections.json. A title you fix only in the browser (without server-admin or an export) shows locally but reverts on the next rebuild, because the build regenerates the catalog from data/.

Manual identification

• Covers that couldn't be read are grouped on identify.html, where you name them (queued for discovery on the next online build) or discard them (e.g. blank tapes).

---

How it compares

Most movie-collection tools add items by barcode scan or manual entry and keep your catalog in their cloud or a dated desktop app. MediaHound is the only one that identifies titles from photos of the covers and generates a modern static website you own and host for free — offline-first and open-source. It's also one of the few that handles VHS (which usually has no scannable barcode in the disc databases others rely on).

	MediaHound	CLZ / Libib	Tellico / GCstar	Plex / Jellyfin
Add by photo of cover	✅ OCR/AI	❌ barcode/manual	❌ search/manual	❌ scans video files
Modern website you host free	✅	❌ their cloud	◻︎ dated HTML export	❌ private server
Open-source / offline / no account	✅	❌	✅ desktop	✅ (Jellyfin)
For a physical shelf	✅	✅	✅	❌ digital files

See COMPARISON.md for the full, honest analysis — including when a barcode app (CLZ/Libib), an OSS desktop cataloger (Tellico/Data Crow), or a media server (Plex/Jellyfin) is the better choice.

---

Providers (how titles get identified & enriched)

Both paths are first-class — pick them per-site in config.toml. The default needs zero keys.

Concern	Default (no key)	Optional upgrade
Identify title from a cover	tesseract — open-source OCR	claude (Anthropic vision, also writes the intro) · ollama (local model)
Movie metadata + poster	wikidata — Wikidata + Wikipedia + Wikimedia	tmdb (free key) · omdb (free key)
Music metadata + cover art	musicbrainz — MusicBrainz + Cover Art Archive	discogs (planned)
Where to watch / listen	justwatch (movies) · keyless Spotify/Apple/YouTube search (music)	Spotify / Apple Music keys (planned)
Resale	eBay sold-listings link + estimate	Discogs price (planned, music)

Switch to a premium provider in config.toml:

[identify]
provider = "claude"      # needs ANTHROPIC_API_KEY
[metadata]
provider = "tmdb"        # needs TMDB_API_KEY (or use "omdb" + OMDB_API_KEY)

…and create a gitignored .env next to config.toml:

ANTHROPIC_API_KEY=sk-ant-...
TMDB_API_KEY=...

Robustness built in: results are cached (data/.metadata-cache.json) so rebuilds never re-hit a rate-limited free key, providers fail soft (a bad lookup never drops a title), and a fuzzy match that returns the wrong film is rejected so it can't corrupt your names.

---

Offline by default

mediahound build is offline — it regenerates the site from existing data and never contacts the internet. Add --online to allow identification / metadata / where-to-watch lookups:

mediahound build --config mysite/config.toml              # offline: just rebuild the site
mediahound build --config mysite/config.toml --online     # online: identify + enrich new titles
mediahound build --config mysite/config.toml --online --refresh-streaming   # also re-check where-to-watch

Useful flags: --mock (demo data), --force (reprocess everything), --limit N, --reidentify <sha256>.

---

Deploy

The generated site folder (mysite/) is plain static files (index.html, identify.html, assets/, data/, posters/, originals/). It's just static files, so you can host it free on GitHub Pages, Cloudflare Pages, Netlify, Vercel, Render, or Surge.sh — no server, database, or build step required. Quickest:

cd mysite && npx netlify deploy --prod          # Netlify
cd mysite && npx wrangler pages deploy .         # Cloudflare Pages
cd mysite && npx surge .                         # Surge.sh

See DEPLOYMENT.md for the full free-hosting comparison plus GitHub Pages, Vercel and S3 instructions. The live demo above is itself hosted free on GitHub Pages via a workflow.

It even works by double-clicking index.html — the build embeds the catalog in data/bundle.js so it loads without a web server.

---

Architecture

See ARCHITECTURE.md for the full picture. In short:

RawImages/*.jpg ─▶ identify (OCR / vision) ─▶ enrich (poster, genres, cast, studio, rating)
                                                    │
                          confidence too low? ──────┴─▶ data/unidentified.json → identify.html
                                                    ▼
   + intro + resale + where-to-watch ─▶ data/collection.json ─▶ static site (index.html)

Python CLI (mediahound/) builds the data; a dependency-free vanilla-JS frontend (mediahound/web/) renders it.

---

Attribution & licensing

• Code: MIT (see LICENSE).
• Default data: Wikidata (CC0), Wikipedia text (CC BY-SA), images via Wikimedia Commons.
• If you enable TMDB: uses the TMDB API but is not endorsed or certified by TMDB.
• Where-to-watch data via JustWatch; resale links to eBay sold listings (estimates are heuristics).

Security

MediaHound output is a static site — read-only to everyone, with no server to attack. The admin password is a convenience gate, not an access-control boundary (the published catalog can't be changed without rebuilding + redeploying). API keys stay in a gitignored .env; only the password hash ships. All rendered data is HTML-escaped and links are scheme-restricted. See SECURITY.md for the full threat model and reporting instructions.

Contributions welcome — see CONTRIBUTING.md.