v0.11.0「切完就能发」/ "Ready to post"

v0.11.0「切完就能发」/ "Ready to post"

2026-08 生态调研的结论落成两条主线:切片手的时间大头在「切完之后」(多平台规格适配、多账号差异化),而成片观感的差距在声音层。这版把两头一起补上。

📦 平台发布包

导出后自动按平台整理齐套素材到「发布包/」——每平台一个文件夹,拿起来就能发:

• 视频:硬链接,发 N 个平台不占 N 份磁盘
• 封面:按平台画幅重裁(小红书 3:4、B站 16:10、视频号 6:7、抖音/快手竖版),上偏构图保人脸
• 文案:按平台上限适配——小红书标题硬截 20 字、话题数按各平台建议值收敛
• manifest.json 记录每一处适配(截了哪条标题、裁了什么画幅),适配透明可审计

同一批片发抖音+小红书+视频号,不用再逐平台手改规格。

🃏 一片多版

同一切片一次出 2-3 版差异化包装:不同钩子角度的贴片标题、开场悬念句、发布文案各成一套,封面自动抓下一个响度峰错开帧。

多账号分发靠真差异——2026 年起平台判原创看「实质增量价值」,抽帧/镜像/变速那类像素级去重已被明确判搬运,这条路不做。变体在 clips.json 标 variantOf 可追溯,精华合集与 EDL 自动只收原版。

🔔 音效打点与 BGM

人工剪辑的「声音层」自动化了,规则完全按 2026 年留量剪辑工艺的口径:

• whoosh 卡多段拼接/高潮前置的硬切帧、**「叮」卡全片情绪最高点、开场钩子上屏轻「啵」**一下——每条最多 3 个,宁缺毋滥,音效之间保持最小间距
• 音效由 ffmpeg 本地合成,零素材文件零版权风险;想换真实音效包,替换同名 wav 即可
• BGM:选个本地音频,自动循环铺满全片、音量压在人声之下 17dB、说话时 sidechain 自动闪避、结尾淡出(请用有授权的音乐)
• 混音是独立后处理趟:视频流直拷,画质零损失;质检在其后照常复核最终混音

✂️ 剪得更懂节奏

• 停顿阈值按品类分档:赛事解说 0.4s、口播 0.6s、访谈对谈 0.9s——对谈的呼吸感不再被剪成机关枪
• 笑点前的停顿禁删:情绪峰值事件前后 1 秒的空隙是节目效果(抖包袱前的憋),跳剪不再「优化」掉它
• 自动运镜绑定真实事件:推近时刻对准情绪峰值,不再只是匀速呼吸
• 质检新增节奏告警:无字幕无运镜的片子,超过 5 秒无视觉变化会在 qa 里提醒
• 新字幕样式**「动态极简」**:短块卡点上屏、白字细描边软阴影、每块至多一处品牌色高亮(关键词/价格数字优先)——Hormozi 大字疲劳后的 2026 主流审美

⬆️ 升级方式

到 Releases 下载新安装包覆盖安装即可。模型、词表、偏好原位保留。

---

Two takeaways from our 2026 ecosystem research shaped this release: clippers now spend most of their time after the cut (per-platform specs, multi-account variants), and the perceived quality gap lives in the sound layer. This version addresses both.

📦 Platform publish packs

After export, assets are organized into a ready-to-post folder per platform:

• Video: hardlinked — posting to N platforms doesn't cost N copies of disk
• Covers: re-cropped per platform (RedNote 3:4, Bilibili 16:10, WeChat Channels 6:7, vertical for Douyin/Kuaishou), biased upward to keep faces
• Copy: adapted to platform limits — RedNote titles hard-truncated at 20 chars, hashtag counts per platform
• manifest.json records every adaptation, so nothing changes silently

🃏 One clip, multiple versions

Each clip can ship as 2-3 differently packaged versions: different hook-angle title cards, opening teasers and post copy, with covers pulled from different loudness peaks.

Multi-account distribution through genuine differentiation — platforms now judge originality by "substantive added value", and pixel-level dedup tricks (frame drops, mirroring, speed shifts) are explicitly flagged as reposts, so we won't build them. Variants carry variantOf in clips.json; compilations and EDL include originals only.

🔔 SFX cues & BGM

The sound layer of a human edit, automated with rules straight from 2026 retention-editing craft:

• A whoosh on stitch/cold-open hard cuts, a ding on the clip's emotional peak, a soft pop as the opening hook lands — at most 3 per clip, spaced apart, less is more
• Effects are synthesized locally by ffmpeg — zero assets, zero licensing; drop in same-named wav files to use your own pack
• BGM: pick a local audio file — looped to fit, kept 17dB under the voice, sidechain-ducked while speaking, faded out at the end (use licensed music)
• Mixing is a separate post-pass: video stream copied untouched, and QA re-verifies the final mix

✂️ Smarter pacing

• Pause thresholds by genre: 0.4s for esports casting, 0.6s for solo talk, 0.9s for interviews — conversations keep their breathing room
• Protected pauses: gaps within 1s of an emotional peak (the beat before a punchline) are never jump-cut away
• Auto-zoom tied to real events: push-ins now land on emotional peaks instead of a fixed breathing rhythm
• New QA pacing check: clips with no captions and no zoom get a warning when nothing changes visually for over 5 seconds
• New dynamic-minimal caption style: short punch-in chunks, clean white with soft shadow, at most one brand-color highlight per chunk (keywords/prices first)

⬆️ Upgrading

Grab the new installer from Releases and install over the old version. Models, glossary and preferences stay in place.