The AI editor that actually sits through your replays

What it does

OpenClip ingests long videos—livestream replays, podcasts, and talking-head broadcasts—from Bilibili, YouTube, or local files. It transcribes, segments, and runs an LLM over the content to extract the most engaging moments as standalone clips, complete with auto-generated titles, covers, and subtitles. The project bills itself as a lightweight alternative to AutoClip, focused on automation rather than manual editing.

The interesting bit

The pipeline automatically routes transcription by language—Whisper for English, Paraformer for Chinese—and can tag speakers by name in interviews or debates. A natural-language --user-intent flag lets you steer the LLM toward specific topics, so you are not stuck with whatever the algorithm thinks is “viral.”

Key highlights

• Three interfaces: a Streamlit web UI, a CLI, and an Agent Skill for Claude Code and TRAE that accepts plain-English commands.
• Optional --deep-optimize mode runs a second LLM pass to refine clip boundaries and improve how segments stand alone.
• Supports cookie-based downloads from Bilibili and YouTube, multi-P Bilibili videos, and a LAN mode for shared machines.
• Can burn hardcoded subtitles into clips, including bilingual tracks translated through the configured LLM provider.
• Background task queue in the Streamlit UI handles concurrent processing and retries without blocking the browser.

Caveats

• The repository rewrote its Git history in March 2026 after a failed size-reduction attempt; existing clones may need a force reset to recover.
• Speaker identification is explicitly labeled a preview feature, and several capabilities (Paraformer ASR, speaker diarization) require optional extra dependencies.
• YouTube downloads may need a separate JavaScript runtime installed for reliable cookie handling.

Verdict

Livestream operators, podcast producers, and content teams who need to batch-generate social clips from long footage will save hours. If your videos are already short or your brand demands pixel-perfect manual cuts, look elsewhere.