srt-from-script

name	srt-from-script
description	Build an SRT subtitle file from a 3-shot short-drama script (ai-video-script OUTPUT FORMAT). Reads each SHOT_N block's DURATION_S + VOICEOVER, emits cumulative-timestamped SRT cues. Pure text-processing, no LLM, no network. Used by meta-short-drama between merge and the final subtitle-burn step.
description_zh	从3镜头短剧脚本（ai-video-script输出格式）构建SRT字幕文件。读取每个SHOT_N块的DURATION_S和VOICEOVER，输出带累计时间戳的SRT字幕。纯文本处理，无LLM、无网络。由meta-short-drama在合并与最终字幕烧录步骤之间使用。
user-invocable	false
disable-model-invocation	true
provenance	{"origin":"opensquilla-original","license":"Apache-2.0"}
metadata	{"opensquilla":{"risk":"low","capabilities":["filesystem-write"],"requires":{"anyBins":"[Truncated]"}}}
entrypoint	{"command":"python {baseDir}/scripts/build_srt.py","args":["--output","{{ with.output_path }}","--gap-ms","{{ with.gap_ms | default(200) }}","--leading-offset-ms","{{ with.leading_offset_ms | default(0) }}"],"stdin":"{{ with.script }}","parse":"text","timeout":30}

srt-from-script

Parses an ai-video-script 3-shot script and writes an SRT subtitle file whose cues track the script's VOICEOVER per shot, time-coded with cumulative shot durations.

Inputs (with:)

key	required	default	notes
script	yes	—	Full script text (the entire OUTPUT FORMAT block, including OVERVIEW + SHOT_1..N). Passed via stdin so the orchestrator does not need to write a temp file.
output_path	yes	—	Output .srt path. Parent dir created if missing.
gap_ms	no	200	Tail pad subtracted from each cue's end so the subtitle vanishes ~200 ms before the next shot starts — avoids cuts clipping mid-character.
leading_offset_ms	no	0	Shifts every cue forward by this many ms. Set to the cover/intro clip duration when the merged video prepends a title card before SHOT_1.

Parsing rules

• A shot with VOICEOVER: none or empty contributes no SRT cue but its DURATION_S still advances the timeline cursor.
• Cue language follows the script verbatim. Chinese stays Chinese, English stays English — no translation.
• Cumulative timestamps: SHOT_1 starts at 00:00:00,000; SHOT_2 starts at SHOT_1.duration; etc.
• End time of each cue = next-shot start − gap_ms, clamped to ≥ 800 ms after start so very short voiceover lines remain readable.

Output

Prints the absolute path of the written .srt on stdout. The file is UTF-8 encoded so CJK voiceover lines survive when ffmpeg reads them via the subtitles= filter.

Limits

• Assumes the script follows ai-video-script's strict OUTPUT FORMAT (=== SHOT_N === blocks with DURATION_S: and VOICEOVER: fields). Drift away from that format → zero cues, exit 1.
• 3-5 shots tested. Larger shot counts work but timestamps grow.