Skip to content

Netflixy Main Subtitle Intent Pack

This pack targets practical streaming efficiency while preserving viewer-intent essentials.

Intent

This pack standardizes mixed-library inputs into streaming-friendly HEVC outputs while preserving mainline viewing intent.

  • preserve full audio when possible
  • preserve director-intended english "main subtitle" behavior when detected
  • use MKV when subtitle intent applies, otherwise emit stream-ready fragmented MP4

What It Optimizes For

  • practical "netflixy" bitrate reduction posture
  • consistent HEVC delivery across 4K, 1080p SDR, and legacy sub-HD lanes
  • audio preservation first, without forcing broad audio transcoding
  • subtitle-intent-sensitive container selection
  • optional legacy remediation (deinterlace + stable black-bar crop) in the sub-HD lane

Guardrails

  • 1080 profile lane is SDR-gated (bt709) in the 1280x720..1920x1080 envelope.
  • 4K profile lane accepts SDR or HDR candidates in the configured 1920x1080..3840x2160 envelope.
  • Legacy sub-HD profile lane accepts broad codec/color intake in the 320x240..1279x719 envelope.
  • Codec intake is broad across lanes (any), so HEVC, H.264 (including rare 10-bit), AV1, VP9, and legacy MPEG-2 style mezzanines can be processed.
  • Guardrail misses are written as *.guardrail_skipped.txt markers by profile_guardrail_skip.sh.

Focus

  • preserve all audio streams
  • preserve one "main subtitle" when it appears director-intent oriented
  • when subtitle intent is present, emit MKV for robust subtitle compatibility
  • when subtitle intent is absent, emit stream-ready MP4 (fragmented + init/moov-at-start by default)
  • for legacy sub-HD lane: optional interlace-aware deinterlace + stable black-bar auto-crop

Included Profiles

Pack Flow

flowchart LR classDef stage fill:#e0f2fe,stroke:#0284c7,color:#0c4a6e,stroke-width:1.2px; classDef gate fill:#fff7ed,stroke:#f59e0b,color:#7c2d12,stroke-width:1.5px; classDef output fill:#dcfce7,stroke:#16a34a,color:#14532d,stroke-width:1.2px; classDef skip fill:#f3f4f6,stroke:#6b7280,color:#1f2937,stroke-width:1.2px; subgraph Intake["Lane Classification"] A[Candidate media]:::stage --> B{Profile envelope match}:::gate B -->|4K 1920x1080..3840x2160| C[4K lane]:::stage B -->|1080 SDR bt709 1280x720..1920x1080| D[1080p lane]:::stage B -->|Legacy sub-HD 320x240..1279x719| E[Legacy lane]:::stage B -->|No match| X[Routed elsewhere or guardrail-skipped]:::skip end subgraph Processing["Per-Lane Processing"] C --> C0[Run subtitle-intent action]:::stage D --> D0[Run subtitle-intent action]:::stage E --> E0[Optional deinterlace and stable auto-crop]:::stage E0 --> E1[Run subtitle-intent action]:::stage end C0 --> Cq{Main subtitle intent detected?}:::gate D0 --> Dq{Main subtitle intent detected?}:::gate E1 --> Eq{Main subtitle intent detected?}:::gate Cq -->|Yes| Cmkv[Emit MKV preserving subtitle intent]:::output Cq -->|No| Cmp4[Emit fragmented MP4 stream-ready]:::output Dq -->|Yes| Dmkv[Emit MKV preserving subtitle intent]:::output Dq -->|No| Dmp4[Emit fragmented MP4 stream-ready]:::output Eq -->|Yes| Emkv[Emit MKV preserving subtitle intent]:::output Eq -->|No| Emp4[Emit fragmented MP4 stream-ready]:::output

What This Pack Does Not Do

  • It does not normalize frame rate; source cadence/timebase is preserved by default.
  • It does not transcode audio for target-device compatibility by default.
  • It does not guarantee every mezzanine audio codec can be muxed into every output container.
  • It does not semantically understand subtitle meaning; subtitle selection uses metadata/flag heuristics.
  • It does not OCR or convert bitmap subtitles to text subtitles.
  • It does not generate ABR ladders (HLS/DASH); outputs are single-file delivery artifacts.
  • It does not certify playback on every device model; profiles are compatibility-oriented guardrails.
  • It does not enforce PSNR/SSIM/VMAF thresholds unless quality checks are explicitly enabled and configured.
  • It does not invent missing HDR/DV essence; metadata repair is heuristic and can be disabled.
  • It depends on source integrity and toolchain support for DV/HDR retention; strict mode may fail instead of silently downgrading.