Nishi Family › Compare › Media Serving, Management and Delivery
Nishi vs the Field
Media Serving, Management and Delivery — mechanically measured, liar-killed, sovereign.
Nishi /gallery + /library vs Jellyfin and Plex and Immich and Kavita
How this is scored. Every Nishi cell is measured: the generator reads the real organ source on disk and requires the implementing symbol to exist (no self-grading). Competitor cells record documented capability presence. Best=leads this axis, Yes=present, Part=partial, No=absent. This is capability presence, not depth or scale: the majors lead on index size and neural ranking. Nishi's genuine exceeds are the sovereignty / neutrality / determinism axes.
Feature matrix
| Capability | Nishi | Jellyfin | Plex | Immich | Kavita | Notes |
|---|---|---|---|---|---|---|
| Analysis-on-ingest derivation pipeline | Yes | Best | Best | Best | Yes | LIVE 2026-07-14: one idempotent pass derives thumbnails + NXVI on ingest, auto-fired per torrent; incumbents scan-on-add with deeper metadata agents |
| Image thumbnail generation + cache | Yes | Yes | Yes | Yes | Yes | Cache-first thumbs, cold-generate via nx_galx_thumb; all incumbents thumbnail on ingest |
| Video poster / recording thumbnail | Yes | Best | Best | Yes | No | Poster.jpg served instantly for cam recordings; harvested video is a gap (no frame extractor yet); incumbents grab a frame on scan |
| Duration + keyframe index (seekable MSE) | Yes | Best | Best | No | No | Sovereign NXVI keyframe index for accurate seek; H.264 only (HEVC gap); incumbents probe all codecs |
| On-the-fly remux + MSE streaming | Yes | Best | Best | No | No | TS->fMP4 fly-remux + MSE segments; no adaptive ladder yet; incumbents ship HW-transcode ABR ladders |
| Dead-air / motion / scene / rhythm markers | Best | Part | Part | No | No | Sovereign demux-signal analyzer emits NXMK markers on the seekbar; Jellyfin/Plex only do intro-skip, no motion/rhythm classification |
| On-demand super-resolution upscale | Best | No | No | No | No | Sovereign 2x upscale endpoint per asset; no incumbent media server upscales media on demand |
| Auto-derive on ingest (orchestrated) | Yes | Best | Best | Best | Yes | Torrent worker fires the guarded galxpipeline sub after indexing; incumbents auto-scan watch folders |
| Unified media library listing + filters | Yes | Best | Best | Best | Best | Mixed image+video grid, state filters, BM25 search; incumbents have mature library UIs |
| Book / manga / comic reader library | Yes | No | No | No | Best | /library serves books/manga/photos/STL; Kavita is a dedicated manga/book server and leads depth |
| Range-streaming file delivery | Yes | Yes | Yes | Yes | Yes | HTTP Range on any NAS file; all incumbents range-stream |
| Book cover thumbnail extraction | Yes | No | No | No | Best | Embedded-cover extract + cache; Kavita/Komga extract covers on scan |
| LLM-native content packing (push a book into an LLM) | Best | No | No | No | No | /api/llmpack packs a book into a token budget natively; no media server does this |
| Per-user auth + private library area | Yes | Yes | Yes | Yes | Yes | Modern auth + per-user private area; all incumbents are multi-user |
| Resilient multi-strategy fetch (acquisition) | Best | No | No | No | No | Cascade + classify + always-inform fetch for INGEST sources; incumbents lean on external download clients |
| Trickplay / BIF scrub-preview sprites | No | Best | Best | No | No | Top delivery-UX gap; Jellyfin .trickplay + Plex BIF generate scrub-preview sprites on scan |
| Chapters + subtitle / audio-track selection | No | Best | Best | No | No | No chapter markers or track menus; incumbents parse container chapters + expose track selection |
| Face / object / CLIP semantic media search | No | No | No | Best | No | Immich leads: on-device face clustering + CLIP semantic photo search; ours is text BM25 (pose/ViTPose port exists but is unwired to serving) |
| HEVC / broad-codec playback (transcode) | No | Best | Best | Yes | No | The genuine won't-play gap; incumbents HW-transcode HEVC/AV1/VP9 to browser-safe streams |
| Adaptive-bitrate HLS / DASH ladder | No | Best | Best | No | No | No ABR ladder; incumbents ship multi-rendition HLS/DASH for bandwidth adaptation |
| Comic / manga page rendering (CBZ/CBR correctness) | No | No | No | No | Best | OPERATOR BUY-OFF 2026-07-14: comic/manga rendering is BROKEN in practice -- the nx_cbx_pages symbol exists but the served format is wrong, so symbol-presence over-credited it; Kavita/Komga render CBZ/CBR reliably |
| Polished easy-on-the-eyes UI (light + dark theme) | No | Best | Best | Best | Best | OPERATOR BUY-OFF 2026-07-14: the screen is a harsh plain white with no dark theme -- not easy on the eyes; Jellyfin/Plex/Immich/Kavita all ship polished dark + light themes |
| Responsive / non-clunky UX polish | No | Best | Best | Best | Yes | OPERATOR BUY-OFF 2026-07-14: the UI is clunky; incumbents have refined, responsive, mobile-first interfaces |
| Accessibility (contrast / keyboard / ARIA) | No | Yes | Yes | Yes | Yes | No measured contrast/keyboard/ARIA pass; incumbents meet baseline accessibility |
| Live UX-judged quality (beyond symbol-presence) | No | Best | Best | Best | Best | MEASUREMENT DISCIPLINE: a symbol on disk is NECESSARY-NOT-SUFFICIENT. UX polish, format correctness and theme are NOT captured by c_file_has -- they need a live UI-judge + user buy-off. Operator flagged clunky/broken/white -> the earlier 750 was over-optimistic; this axis exists so the census stops over-crediting presence |
| Sunlight / outdoor readability | No | Yes | Yes | Yes | Yes | COMPLETENESS-ADVERSARY MISS caught 2026-07-14 (operator flagged): a mobile web reader needs a high-contrast/auto sunlight mode -- e-ink wins outright; incumbents' native apps offer high-brightness; ours has none |
| Ambient-light auto-adaptation | No | Yes | Yes | Yes | Yes | COMPLETENESS MISS: no ambient-light / auto-brightness / time-of-day adaptation; native reader apps adapt to sensor + time |
| Mobile / responsive legibility | No | Best | Best | Best | Best | COMPLETENESS MISS (operator flagged): the reader is not tuned for small-screen one-handed reading; incumbents ship mobile-first apps |
| Anti-glare / matte reading surface | No | Yes | Yes | Yes | Yes | COMPLETENESS MISS: no matte/low-glare treatment (pure-white glare is the harsh case); reader apps offer paper/sepia low-glare modes |
| Color temperature / warm night shift | No | Yes | Yes | Yes | Best | COMPLETENESS MISS: no warm color-temperature or night shift; Kavita/Kindle have warm + sepia reading modes |
| Flicker-free / refresh comfort | No | Yes | Yes | Yes | Yes | COMPLETENESS MISS: no flicker/refresh consideration; e-ink is flicker-free by nature |
| Reflective / paper-like feel | No | No | No | No | Yes | COMPLETENESS MISS: emissive-only; the researched comfort target is a reflective paper-like feel (e-ink) -- a category-wide web gap |
| Reading typography (measure / leading / serif) | No | Yes | Yes | Yes | Best | COMPLETENESS MISS: no reading-optimized typography (serif long-form, ~65ch measure, leading); reader apps tune all three |
| Blue-light reduction / circadian | No | Yes | Yes | Yes | Best | COMPLETENESS MISS: no blue-light reduction / circadian awareness; reader apps + OS night modes reduce blue light |
| Resume / watch-state per user | No | Best | Best | Best | Yes | COMPLETENESS MISS: reader /api/pos position-sync exists but resume/watch-state is not unified across the media plane; incumbents track it everywhere |
coverage 428/1000EXCEEDS 4PRESENT 11ABSENT 20
Honest verdict. Nishi has broad, sovereign search-engine presence — a real organ behind every core ranking and crawl axis, built bits-up in one language. It is not yet a scale or neural-ranking rival: the majors lead on index size (billions to trillions vs a Common-Crawl-seeded slice) and trained rerankers. Nishi's genuine exceeds are structural — owning the whole stack, neutrality by construction, and a zero-JS integer-deterministic result page. The climb is a bigger index and a trained reranker.
Generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/media.matrix — every Nishi cell verified against real organ source on disk. Zero JS, zero trackers.