Opus Clip, but it never phones home
HotClip is a free, fully local desktop app that turns long videos and livestream VODs into vertical short clips — with danmaku as evidence for what counts as a highlight.

What it does HotClip is an Electron app (Windows/macOS, Linux experimental) that ingests long videos, livestream replays, podcasts, or pasted B站/YouTube URLs, transcribes them locally, and picks out highlight segments with a “viral score,” opening hook, and a stated reason for every cut. It then exports 9:16 vertical shorts with dynamic word-by-word subtitles, cover images, and per-platform publishing copy for Douyin, Kuaishou, Bilibili, Xiaohongshu, 视频号, and TikTok. AGPL-3.0, no watermark, no credits, no registration, no uploads.
The interesting bit It folds danmaku (live chat) files from B站录播姬 and Douyin recorders directly into highlight detection — weighting gifts, SCs, and likes while filtering single-person spam. The audience voted second-by-second on what was actually good, and most clip tools ignore that evidence entirely. Also unusual: every AI cut comes with its reasoning and word-aligned timestamps, plus an independent sub-1MB voice-activity check so cuts don’t clip speech trailing into pauses.
Key highlights
- Fully local pipeline: transcription (SenseVoice / Paraformer / FireRedASR2), clipping, subtitles, and export never leave your machine; optional cloud engines are explicitly labeled by privacy level.
- Danmaku heat curves drawn on the timeline, with candidate segments landing on the peaks.
- Watch folders for OBS/录播姬 directories — it clips replays automatically as they finish, 7×24, with a persistent task queue that survives restarts.
- CLI and MCP server (
clip_video,detect_highlights) so agents can drive it; genre, clip length (10s–120s), and max clip count are all parameterized. - Post-publish learning: import playback/interaction CSVs and it reconciles against stable content IDs, running local A/B comparisons across multiple versions of the same clip.
- Sensible engineering details: atomic per-encode file replacement, render cache reuse, HDR-to-SDR conversion for phone footage, undo/redo history that persists across project reopen.
Caveats
- Windows builds are unsigned (SmartScreen warning) and macOS builds are ad-hoc signed, not notarized — expect first-launch friction.
- Linux is labeled experimental and may need
--no-sandbox. - The optional smart voice-enhancement model (~10MB) downloads on first use; it falls back to the basic filter if unavailable.
Verdict Worth a look if you cut Chinese-platform livestream content and resent credit-based SaaS clip tools — the danmaku-driven detection alone is a genuinely different idea. If your workflow is English-only YouTube Shorts with no chat data, the core value proposition thins out, though the local-first pipeline still holds up.
Frequently asked
- What is xixihhhh/hotclip?
- HotClip is a free, fully local desktop app that turns long videos and livestream VODs into vertical short clips — with danmaku as evidence for what counts as a highlight.
- Is hotclip open source?
- Yes — xixihhhh/hotclip is open source, released under the AGPL-3.0 license.
- What language is hotclip written in?
- xixihhhh/hotclip is primarily written in TypeScript.
- How popular is hotclip?
- xixihhhh/hotclip has 503 stars on GitHub.
- Where can I find hotclip?
- xixihhhh/hotclip is on GitHub at https://github.com/xixihhhh/hotclip.