@remixmate/cli 0.9.37 → 0.9.38

package/dist/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.9.37",
-  "generatedAt": "2026-09-21T03:49:51.210Z",
+  "version": "0.9.38",
+  "generatedAt": "2026-09-21T11:34:27.703Z",
   "skills": [
     {
       "id": "export-jianying",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remixmate/cli",
-  "version": "0.9.37",
+  "version": "0.9.38",
   "description": "AI media generation skills for Claude Code / Codex — 12 skills covering image, video, voice, digital human, web screenshot, web recording, script, template registry, rendering, Jianying export, and video deconstruction.",
   "type": "module",
   "main": "dist/index.js",

package/skills/gen-script/SKILL.md

@@ -441,11 +441,22 @@ Apply the following principles when producing the DSL:
 1. **Narrative-first**: opening hook → point expansion → CTA close.
 2. **Assets are declared, not executed**: every asset starts as `AssetRef` with `status: planned`; no generation API call from this skill.
 3. **Narration drives duration**: each scene's duration should match its narration reading time (≈ 3–4 zh chars/sec, or ~2 words/sec for English).
-4. **Segment narration at subtitle granularity**: separate segments inside `audio.narration.text` with `\n`; each segment is 10–16 Chinese characters (or an equivalent number of English words). TTS returns per-line timestamps split on `\n`; the renderer uses them directly as subtitle cut points. Segments longer than 18 characters wrap awkwardly on screen; segments shorter than 6 flicker. Example:
+4. **One complete thought per line — do NOT pre-split at subtitle width.** Separate lines inside `audio.narration.text` with `\n`, and put **one full clause or sentence** on each line (roughly 15–30 Chinese characters, or an equivalent number of English words). A `\n` is a hard cut in the *audio*: TTS treats each line as its own segment and returns a timestamp for it, so lines should fall where a human would actually pause for breath.
+
+   **Do not cut mid-phrase to make lines "subtitle-sized".** The renderer splits each line into on-screen subtitles itself, and it does that better than you can from the text alone: it measures the real glyph widths for the actual font size and frame (`HNSW` is not as wide as four Chinese characters), knows how many characters fit on one line of *this* template and aspect ratio, and picks cut points by punctuation strength (sentence end → colon → comma → enumeration comma → space). It also guarantees no wrapping. Pre-splitting takes all of that away and just makes the captions choppy.
+
+   Measured on a shipped video (2026-09-21): narration written as 6–8 character lines produced 57 subtitle screens in 121.5s — a median of 2.10s each, 33% of them under 1.5s, with `AI Agent` alone on a screen. The renderer's own splitter never even fired.
+
+   Write:
    ```
-   "text": "Alibaba's open-source AI video-editing project\nalready has 4.3k stars on GitHub"
+   "text": "最近刷爆科技圈的 AI Agent 到底厉害在哪\n真正的分水岭不是文案更精准，而是自主执行能力"
    ```
-   instead of one continuous paragraph.
+   not:
+   ```
+   "text": "最近刷爆科技圈的\nAI Agent\n到底厉害在哪\n不是文案更精准\n也不是对话更流畅"
+   ```
+
+   **Use full-width punctuation in Chinese narration** (`，。：；？！`), not half-width (`,.:;?!`). Half-width punctuation in Chinese text renders with no trailing space — `第一步,目标感知:只给最终目标` reads cramped on screen.
 5. **Moderate scene count**: 30-second videos work well with 4–6 scenes, 60-second videos with 6–10.
 6. **Leave room for templates**: pick generic layouts; do not assume a specific template implementation.
 7. **Image model allowlist**: every `type: image` + `source: gen-image` `AssetRef`'s `payload.model` **must** be one of the values in the table below. **Never** use display names, short forms, or made-up ids (e.g. `seedream`, `seedream-5`, etc.).

package/skills/template-registry/video_dsl/runtime/timeline_compiler.py

@@ -20,10 +20,35 @@ from typing import Optional
 CHARS_PER_SECOND = 5.0
 
 _PRIMARY_BREAK = re.compile(r"(?<=[。！？；\n])")
-_SECONDARY_BREAK = re.compile(r"(?<=[，,、：:])")
+# 次级断点**不含顿号 `、`**。顿号是并列项之间的分隔，在它上面下刀就是把一串并列
+# 拦腰切开（`调研、策划、` ／ `开发、分析，…`）。这一层只数字符、分不出标点的语义
+# 强度，所以干脆不碰它 —— 真需要再切时交给渲染端，那边是分级的（句末 > 冒号 >
+# 逗号 > 顿号 > 空格），只有前面几级都切不动才会降到顿号。
+_SECONDARY_BREAK = re.compile(r"(?<=[，,：:])")
 _PRIMARY_TAIL = re.compile(r"[。！？；\n]\s*$")
 
-MAX_SEGMENT_CHARS = 16
+# ─── 这两个上限管的是「喂给 TTS 的行」，不再是「一屏字幕」 ────────────────
+#
+# 曾经是 16 / 6，那是渲染端还不会分屏时的产物：谁都不会再切，所以必须在这里
+# 就把旁白预切成一屏放得下的大小。现在渲染端（core 的 `chunkSubtitleSegments`）
+# 会按**实际字号与实际可用宽度**算容量、按**标点语义强度分级**切分，这件事它
+# 做得比这里好得多：
+#   · 这里只能数字符数，数不出 `HNSW` 比 4 个中文字窄一半；
+#   · 这里的 `_SECONDARY_BREAK` 是平的，冒号和顿号同权，会把并列项拦腰切开；
+#   · 这里一个全局 16 要同时服务 16:9（一行放 44 字）和 9:16（一行放 22 字）。
+#
+# 于是 16 的唯一效果变成了**抢在渲染端之前把句子剁碎**。2026-09-21 线上成片实测：
+# 121.5s 里 57 屏、中位 2.10s 一屏、33% 短于 1.5s，「AI Agent」独占一屏 —— 全都是
+# 这个 16 的产物，渲染端的分屏阈值（20）一次都没被触发。
+#
+# 现在这里只负责一件事：**在真实语音停顿处断行**，让 Minimax 按停顿返回时间戳。
+# 句末（`_PRIMARY_BREAK`）永远断；逗号那一级只在句子长到一行 TTS 太长时才用。
+# 30 是「一个中文长句」的量级 —— 越过它说明这句本来就该有个呼吸点。
+#
+# ⚠️ 别再把它当字幕宽度调：想改一屏显示多少字，改 core 的
+# `DEFAULT_SPLIT_MAX_UNITS` / `LINE_FILL_LIMIT`，那边才看得见字号和画幅。
+MAX_SEGMENT_CHARS = 30
+# 合并下限保持 6：TTS 行太短会让合成的语气断掉（和字幕闪烁是两回事）。
 MIN_SEGMENT_CHARS = 6
 
 # 字幕显示时需要去掉的末尾标点（中英文句末 + 逗号等）
@@ -157,10 +182,15 @@ def _strip_trailing_punct(text: str) -> str:
 
 
 def segment_narration(text: str) -> list[str]:
-    """Split narration text into subtitle-display-sized segments (text only).
+    """Split narration into **TTS lines**, cutting at speech pauses.
 
     Use before TTS: ``"\\n".join(segment_narration(text))`` sends to Minimax
-    so it returns per-segment timestamps.
+    so it returns per-line timestamps.
+
+    ⚠️ 这里切的**不是字幕屏**。字幕分屏归渲染端（core 的
+    ``chunkSubtitleSegments``），它知道字号、画幅和标点的语义强度；这里只知道
+    字符数。这一层切得越细，留给渲染端的余地越小 —— 见 ``MAX_SEGMENT_CHARS``
+    的注释。
     """
     if not text:
         return []