Clip Finder Pro — Ready-to-Cut Transcript Clips

Real model run (DeepSeek V4.1 Flash, thinking on, 1 call, with a saved channel profile: "public affairs podcast", calm and factual tone, 25-45 s clips): a 90-minute public-domain US House hearing transcript (Rare Earth Minerals and 21st Century Industry, 2010; archive.org ASR captions) → the model picked 8 clips (videos over 1 hour get up to 8) and all 8 were exported, 7 of them inside the preferred 25-45 s. E.g. #1 00:41:41 "None of these technologies will work without rare earths" (score 79, risk medium); #2 00:56:37 "Seventeen scientists competing with six thousand Chinese" (77). Self-check: 3 ready, 4 to review, 1 needs trim; all 8 rated medium risk (verbatim speech, no creator commentary yet). Text is the raw ASR transcript. This is the only example shown here, and it is real model output.

Editor files: Exports a CMX3600 EDL, Final Cut Pro XML (FCPXML 1.9), a Premiere-compatible XML (xmeml v4) and a DaVinci Resolve marker EDL. Files are validated with OpenTimelineIO; direct import into Premiere Pro, DaVinci Resolve and Final Cut Pro has not been tested on every version — if an import fails, the included cut_clips script (ffmpeg) cuts the clips directly. In the Premiere XML each clip is its own clip on the timeline, end to end, with a marker per clip; the Resolve marker EDL marks each clip on a timeline that starts at 01:00:00:00 (colour = status). Relink to your video once.

Cutting script: cut_clips.sh (macOS/Linux) and cut_clips.bat (Windows) cut every clip from your video with ffmpeg on your own computer; add --vertical for a centred 9:16 crop and --subs to burn in captions. The scripts never delete or overwrite files and make no network requests. You need ffmpeg installed.

Vertical-video captions: clip_XX_shorts_captions.srt per clip, timed to the cut clip, short lines (up to 32 English characters or 16 Chinese characters, at most 2 lines).

Clip map: clip_map.html (opens offline in any browser, no external scripts) and clip_map.md: where each clip sits in the full video, score, risk, status, and a copy button per clip.

Channel profile: set your channel type, tone, target platforms, preferred clip length and banned words once; they are saved and applied to every video. A clip whose title or copy uses a banned word is flagged for review. posting_plan.md spreads this month's exported clips over a posting calendar (a suggestion; move posts freely).

Also per-clip SRT files on the full-length timeline and a paste-ready list for CapCut, Opus Clip, Descript or any AI editor.

Explainable scoring: four factors, each with a one-line reason.

Reused-content risk check per clip: flags verbatim cuts, a missing commentary hook and third-party content, and suggests concrete commentary, overlays or edits to add.

Copy pack per clip: title, description and up to 5 hashtags for each platform, checked against length limits.

Upload several transcripts; I process them one per message and keep a running cross-video ranking. Reply "next" (or 继续 / 好) to continue. Files you already processed are skipped. When your queue is empty you get a cross-video bundle: top clips table, one merged EDL/FCPXML and a CSV.

Clip self-check per clip (local rules): hook, complete start, clean ending (catches clips that stop mid-sentence, trail off or end on a dangling question), title/content match, context dependency and risk flags, with a ready / needs trim / review / rework verdict. Clips that need a trim are never marked ready.

Nothing is dropped silently: if a selected clip fails a check, the reply lists it with the reason; everything else is exported.

Raw transcript: clip text is your transcript exactly as uploaded, so speech-recognition (ASR) errors are kept as they are. Fix them in your editor before posting.

Padding: each clip comes with suggested in/out points padded by 1.5 s at each end, so the first and last words aren't cut off.

Platform fit: suggested in/out points for Shorts, Reels and TikTok, snapped to sentence boundaries (rules of thumb, not platform policies).

Monthly insights: which hook styles and topics scored best in your own clips this month.

Try it first: say "demo" to see a finished result from a built-in sample in seconds.

No video upload needed: text only, fast and private. Caption files (.srt/.vtt) exported from your video platform work directly.

Time saved (estimate, not a guarantee): picking clips by hand from a 1-hour episode typically takes 1.5–3 hours; here it takes minutes plus your review.

3-minute quickstart: upload a transcript → open clip_map.html and review the risk check → try clips_premiere.xml / clips_resolve_markers.edl / clips.fcpxml in your editor (relink media); if an import fails, run cut_clips.sh / cut_clips.bat to cut the clips directly.

Compared with doing it yourself for free (your transcript, a media player and your editor; facts only):

Video length: For best results keep each video to about 1.5 hours; the hard limit is 2 hours. 单条视频建议控制在 1.5 小时左右，最长不超过 2 小时。

Billing: US$15/month for up to 60 successful video exports a month and 70 requests in total (both reset on the 1st, UTC). Each video uses 1 request; retries and questions count as requests too (every message counts, including "next"), so the extra 10 can go to retries and questions. Each successfully exported video counts as 1 run (1 video); a failed export doesn't use your video quota but does use a request. Skipped files and duplicate uploads are not counted as videos. Up to 2 hours of transcript per video (about 18,000 English words or 30,000 Chinese characters). Longer recordings: split into parts; each part counts as one video.

FAQ:

• Does it download or render video? No, text only.
• YouTube links? Links are not fetched. Please export the captions in YouTube Studio, then upload the file (请在 YouTube Studio 导出字幕后上传).
• How do I get captions? On YouTube: '...more' under the video > 'Show transcript', copy and paste it here. For your own videos: YouTube Studio > Subtitles > ⋮ > Download (.srt/.vtt). Only clip your own videos or videos you are licensed to use.
• Why is there a typo in my clip text? Clip text is the raw transcript; speech-recognition errors are kept as they are.
• How long can a video be? For best results keep each video to about 1.5 hours; the hard limit is 2 hours. 单条视频建议控制在 1.5 小时左右，最长不超过 2 小时。 Longer files are rejected before any model step and are not counted; split them into parts.
• Languages? English and Chinese transcripts. Say 'demo' to see bundled English and Chinese samples.
• How many messages can I send? Up to 70 requests a month in total; every message counts. Each video uses 1 request (one video per message), so 60 videos leave 10 requests for retries and questions.
• Are views guaranteed? No.
• Monetization? Clips that reuse content without your own commentary or meaningful edits may be ineligible for YouTube monetization ("reused content"), even with permission. Shorts longer than 1 minute with a valid copyright claim are blocked. Use the risk check and add your own commentary.