video-subtitle-extractor
Use Video-subtitle-extractor (VSE) to extract hard-coded video subtitles into SRT/TXT, configure OCR/backends, batch GUI runs, clean subtitle output, and synchronize subtitle timelines.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- video-subtitle-extractor
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Video-subtitle-extractor (VSE) to extract hard-coded video subtitles into SRT/TXT, configure OCR/backends, batch GUI runs, clean subtitle output, and synchronize subtitle timelines.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/development-notes.md1.6 KB
- 参考文档references/installation-and-runtime.md3.2 KB
- 参考文档references/repo-provenance.md1.8 KB
- 参考文档references/repo-routing-metadata.json349 B
- 参考文档references/troubleshooting.md2.5 KB
- 脚本scripts/vse_environment_probe.py2.9 KB
03 SKILL.md 正文
Video-subtitle-extractor (VSE)
Use this repo skill when the task involves VSE / Video-subtitle-extractor, hard-subtitle OCR from videos, VideoSubFinder frame detection, PaddleOCR model selection, GUI batch extraction, subtitle post-processing, or the bundled Sushi subtitle timeline synchronizer.
This is an operating guide, not the VSE application itself. If the user has no VSE installation or source checkout yet, first follow installation and runtime to choose a release, source run, or backend-specific environment.
Route by task
| User intent | Read |
|---|---|
Extract hard-coded text from video into .srt/.txt; plan CLI/source runs; choose fast/auto/accurate; reason about VideoSubFinder and cache/output paths |
extraction-workflows |
| Configure PaddleOCR/PaddlePaddle, language codes, bundled V5 models, CPU/CUDA/DirectML/ONNX acceleration, OCR batches, thresholds, backend probes | ocr-backends |
| Edit typo replacement rules, clean OCR text, keep/delete empty timestamps, generate TXT, debug missing OCR lines, tune confidence/area filters | postprocessing-config |
| Operate the PySide6 GUI, batch videos, draw subtitle areas, bind areas to AB sections, choose output folders, stop/retry tasks | gui-batch-operations |
| Synchronize an existing SRT/ASS subtitle track to another video with the bundled Sushi CLI/API or GUI Sync Timeline page | subtitle-sync |
Minimal runtime shape
VSE is distributed primarily as an application/source tree rather than a pip-installable library with package metadata. Common public entry points are:
- GUI:
python gui.pyfrom a VSE source checkout with GUI dependencies. - Interactive extraction CLI:
python -m backend.mainfrom a VSE source checkout; it prompts for video path and subtitle area. - Sushi timeline synchronization:
python -m backend.sushi --src SOURCE --dst DEST --script SUBTITLE -o OUTPUT.
For a non-interactive or automation task, do not guess missing coordinates or backends. Use the bundled planning/probe helpers:
- scripts/vse_environment_probe.py checks dependencies, backend visibility, and VSE source-layout readiness.
- extraction-workflows/scripts/vse_cli_plan.py prints a safe extraction plan without running OCR.
- ocr-backends/scripts/model_config_probe.py reports bundled V5 language/model mapping from a VSE source tree.
Default decisions
- Prefer Fast or Auto mode first. Use Accurate only when Fast/Auto misses many subtitle intervals and the user accepts much slower processing.
- Treat CPU as the portable baseline. CUDA, DirectML, ONNX providers, ROCm, CoreML, and other accelerators are optional speed/accuracy paths and require backend-specific verification.
- Ask for or derive a concrete subtitle area before running extraction. If the user only knows normalized GUI coordinates, route to gui-batch-operations for coordinate conversion.
- Avoid paths with spaces or non-ASCII characters when troubleshooting VSE runs; the upstream README warns these can cause unknown failures.
Cross-cutting references
- Read installation and runtime for Python version, dependency variants, backend install commands, and launch expectations.
- Read troubleshooting for install/import, model, video, path, backend, GUI, and process-management failures.
- Read development notes only for maintainer packaging/build context; release build scripts are not normal user workflows.
- Read repo provenance before relying on this skill for a newer VSE checkout.
Avoid this skill when
- The user wants audio speech recognition / ASR for videos without visible hard-coded text. Use a speech/ASR workflow instead.
- The user wants generic subtitle editing unrelated to VSE extraction or Sushi synchronization.
- The task is only about a different OCR library without VSE, VideoSubFinder, or VSE's bundled model/layout assumptions.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).