Skip to content
Back to search

clipper

SkillWorks with Claude CodeGitHub

Turns any long stream, podcast or video (YouTube, Twitch or Kick) into ready-to-post vertical clips. Works for GTA 6 and any other game, IRL and Just Chatting streams, podcasts and interviews. It finds the biggest streams on any topic, downloads only the audio, builds a hype curve from loudness, YouTube's most-replayed graph and chat speed, transcribes the hot moments locally with Whisper, picks the best moments with a harsh filter, then cuts 9:16 clips with the face cam on top and the main video underneath (or one full-frame crop when there's no face cam), a 3-second hook and word-by-word captions, plus a gallery page with captions and hashtags. Use it when the user says "set up the clipper" (or "set up the GTA clipper"), "clip this stream", "clip this podcast", "find me today's biggest <topic> streams and clip them", "look at what's working", or asks to make Shorts/TikToks/Reels from any stream, podcast or long video.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/twahanur--gta6-clipper/badge.svg)](/listings/twahanur--gta6-clipper)

HTML

<a href="/listings/twahanur--gta6-clipper"><img src="/api/listings/twahanur--gta6-clipper/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

git clone https://github.com/twahanur/stream-clipper ~/.claude/skills/clipper

No git? Click **Code → Download ZIP** above, unzip it, rename the folder to `clipper` and move it into `~/.claude/skills/`. On a Mac that folder is hidden: in Finder press `Cmd + Shift + G`, paste `~/.claude/skills/` and press Enter.

Then open Claude Code (in the terminal, or the Code tab of the Claude desktop app) and say:

> set up the clipper

Setup takes about 5 minutes the first time and ends with `READY`. Allow the commands when Claude asks.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 137 lines
  1. ---
  2. name: clipper
  3. description: Turns any long stream, podcast or video (YouTube, Twitch or Kick) into ready-to-post vertical clips. Works for GTA 6 and any other game, IRL and Just Chatting streams, podcasts and interviews. It finds the biggest streams on any topic, downloads only the audio, builds a hype curve from loudness, YouTube's most-replayed graph and chat speed, transcribes the hot moments locally with Whisper, picks the best moments with a harsh filter, then cuts 9:16 clips with the face cam on top and the main video underneath (or one full-frame crop when there's no face cam), a 3-second hook and word-by-word captions, plus a gallery page with captions and hashtags. Use it when the user says "set up the clipper" (or "set up the GTA clipper"), "clip this stream", "clip this podcast", "find me today's biggest <topic> streams and clip them", "look at what's working", or asks to make Shorts/TikToks/Reels from any stream, podcast or long video.
  4. ---
  5. # Clipper
  6. You turn a 4-hour stream, a 2-hour podcast or any long video into a handful of clips worth posting. GTA 6 is the headline use case, but nothing here is tied to one game or one kind of video. The scripts do the heavy lifting on the user's computer for free; **your job is the judgment**: picking moments, writing hooks, reading the face cam position off a frame, and being honest about what's good.
  7. Paths used below:
  8. - `SKILL` = this skill's folder (the folder this file is in)
  9. - `PY` = `~/.clipper/venv/bin/python`
  10. - `WORK` = `~/Clipper` (every stream, run and clip lives here)
  11. Always run scripts as `cd "$SKILL/scripts" && "$PY" <script>.py ...`.
  12. ## 0. Setup ("set up the clipper")
  13. 1. Run `bash "$SKILL/scripts/setup.sh"`. It installs uv, Python 3.12, yt-dlp, faster-whisper, ffmpeg and Pillow into `~/.clipper` only, plus the free Whisper models. First run takes a few minutes; later runs are instant.
  14. 2. It ends with `READY`. If it fails, show the user the last lines and fix the cause (usually no internet, or the user declined an install).
  15. 3. Create `WORK/preferences.md` if it doesn't exist, with the default rules from section 3 under a heading "My clip rules". This is the file the weekly review updates.
  16. 4. Tell the user it's ready and the two things they can say next: "clip this stream: <link>" (any stream, podcast or long video) or "find me today's biggest <topic> streams and clip them" (for example GTA 6, Minecraft, a streamer's name or "podcast").
  17. If `PY` doesn't exist when the user asks for anything else, run setup first.
  18. ## 1. Find streams (optional)
  19. `"$PY" find_streams.py --query "<topic> stream" [--when today|week|month] [--max 8] [--min-minutes 45]`
  20. Lists the most viewed long streams or videos for whatever the user asks about, with their length. Build the query from their words: "GTA 6 stream", "Fortnite stream", "Kai Cenat stream", "podcast full episode". If they don't name a topic, ask for one in one line. For podcasts use `--min-minutes 30`. Pick the top 2 or 3 unless the user names streamers. Long VODs (2h+) from big streamers are best. Streams that are live right now are skipped by default; clip them once the VOD is up (on YouTube that's usually right after the stream ends).
  21. ## 2. Listen (one stream at a time)
  22. `"$PY" analyze.py "<url>" [--top 40] [--model small] [--game gta]`
  23. - Downloads only the audio (a few hundred MB, not the video).
  24. - Measures loudness every second, pulls YouTube's most-replayed graph, and on YouTube VODs of live streams reads the chat replay to measure chat speed and reactions ("LMAO", "CLIP IT", "???"). Twitch and Kick get loudness only; say so if asked.
  25. - Runs a free sound classifier (YAMNet, `sound_events.py`) on every second, on every stream: it hears gunfire, explosions, crashes, skids, sirens and helicopters (`action`), screams, shouts and laughter (`reaction`), and plain talking (`talk`). High talk with no action is usually a cutscene or chat reading, and it scores lower. Skip it with `--no-sounds`.
  26. - Turns that into one hype curve, takes the top 40 spikes, and transcribes a window around each one (about 2 minutes on long streams, 1 minute on videos under 30 minutes) with Whisper on the user's computer.
  27. - Gives each window a `words` count of reaction words ("oh my god", "bro what", "no no no") and, when the title names a game, that game's words (`profiles.py`; GTA: cops, wanted, wasted, busted, heist, chase, crash...). Force a list with `--game gta`. Add a new game by adding an entry to `PROFILES`.
  28. - Writes `WORK/streams/<id>/for_claude.md` (what you read, strongest candidates first, each with a `sounds:` line), `candidates.json`, `hype.png` (show it, it's a good screen share) and `meta.json`.
  29. Speed: `--model base` is about 2x faster than `small` with slightly rougher transcripts. `--full` transcribes the entire stream (slow, but useful when a stream is quiet and talky, like a podcast). You can run analyze on several streams one after another in the background.
  30. ## 3. Pick (this is your job, be harsh)
  31. Read every `for_claude.md` for this batch, then `WORK/preferences.md` (the user's own rules, which override the defaults below when they conflict).
  32. For gameplay streams, the high-voltage moments are where `sounds:` shows action (gunfire, explosion, crash, siren) or reaction (scream, shout, laughter) **and** the streamer reacts in words. A window that's all game dialogue with high `talk` and no action is a cutscene: cut it unless the streamer's own reaction is the clip. When a window's transcript is sparse (mostly yelling), re-transcribe just that minute before judging it.
  33. For each candidate, decide KEEP or CUT. **A clip is KEEP only if all three are yes:**
  34. 1. **Would a stranger stop scrolling in 2 seconds?** The first line or the first visible beat has to be the reaction, the shock, the joke. Not the setup.
  35. 2. **Does it make sense with zero context?** No "wait, what happened earlier", no inside jokes, no references to a chat message we can't see.
  36. 3. **Does it pay off inside the clip?** It ends on the punchline, the reveal or the reaction, never halfway through a thought.
  37. Also cut: anything that opens with filler ("uh, so yeah", "okay so"), long silent stretches, the streamer reading donations, sponsor reads, ads, music-only stretches (copyright claims), anything sexual, hateful or targeting a real person, and anything that only works as audio-off gameplay with no reaction. Never pick the same moment twice. Big laughter or screaming with no speech can still be a KEEP if the hype score is high and the window around it explains itself.
  38. Moment types to label: `first-reaction`, `freak-out`, `discovery`, `funny-fail`, `hot-take`, `skill-moment`, `rage`, `wholesome`, `glitch`, and for talk content (podcasts, interviews, Just Chatting) `story`, `advice`, `confession`, `argument`.
  39. For talk content the clip lives in the words, so the first line has to be the hook (a bold claim, a number, a question), and the clip has to finish the thought. Cut anything that needs the question asked five minutes earlier.
  40. Score every KEEP 1 to 10 for how hard it would stop a scroll. Keep only 7+. Aim for about 3 per stream, never pad with weak ones. If a stream has nothing good, say so; that's useful information, not a failure.
  41. Set tight times: start 0.5 to 1 second before the first important word or reaction, end right after the payoff. Usually 15 to 45 seconds, never under 8 or over 90.
  42. Write `WORK/picks.json` (overwrite it for each batch):
  43. ```json
  44. {
  45. "picks": [
  46. {
  47. "source": "<stream id from meta.json>",
  48. "start": "01:23:41",
  49. "end": "01:24:09",
  50. "moment_type": "freak-out",
  51. "score": 9,
  52. "streamer": "Kai Cenat",
  53. "hook": "HE WAS NOT READY FOR THIS",
  54. "title": "Kai Cenat loses it at the GTA 6 police chase",
  55. "caption": "The second he saw the helicopter 💀 Credit: Kai Cenat (link in comments)",
  56. "hashtags": ["#gta6", "#kaicenat", "#gaming", "#shorts"],
  57. "why": "Screams in the first second, makes sense alone, ends on the crash"
  58. }
  59. ]
  60. }
  61. ```
  62. Hook rules: 3 to 7 words, ALL CAPS, the way a person would say it ("HE WAS NOT READY FOR THIS"), never a description ("STREAMER REACTS TO CHASE"), no fake claims about what's in the clip, no emoji (the caption font can't draw them). Titles under 60 characters and lead with the person and the moment. Captions are one honest line plus a credit to the streamer. 3 to 5 hashtags. Write the hook and caption in the same language the streamer speaks (the burned-in captions are in that language too), unless the user asks for another one.
  63. Before writing the file, show the user a short table (time, streamer, type, score, hook, why) and the moments you cut with a one-line reason each. Then continue unless they want changes.
  64. ## 4. Cut
  65. 1. `"$PY" cut.py download` downloads only the minutes each pick needs (plus 1.5 seconds on each side), saves three frames per clip in `<run>/frames/` (`<key>_a.jpg` near the start, `_b` in the middle, `_c` near the end), and prints the run folder.
  66. 2. **Look at all three frames of every clip** (open them with your Read tool). Streams and podcasts often switch layouts (a full-screen camera, a BRB screen, a menu, a cut between two podcast cameras), so check the face cam is in the same place in all three. If the layout changes inside the clip, open `<key>_sheet.jpg` (one frame every 2 seconds, left to right, top to bottom, starting at the clip's start time) to see exactly when it switches, then change that pick's `start`/`end` in `<run>/picks.json` so the whole clip uses one layout and re-download just that pick into the same run: `"$PY" cut.py download --run "<run>" --only 03`. Only use `facecam: null` when the whole clip is a full-screen camera or gameplay with no face cam. Most podcasts and interviews are `facecam: null` with `focus_x` on the person speaking. Then fill in `<run>/layout.json` for each key:
  67. - `"facecam": [x, y, w, h]` as fractions of the image (0 to 1) around the streamer's face cam box, tight to the box edges, or `null` if there's no face cam (full-screen camera or gameplay only).
  68. - `"focus_x"`: where the action is horizontally in the gameplay (0 left, 0.5 center, 1 right). For a full-screen camera with no gameplay, set it to the person's horizontal center and keep `facecam` null.
  69. Example: a face cam in the bottom-right corner of a 1920x1080 frame at pixels (1450, 700) to (1900, 1060) is `[0.755, 0.648, 0.234, 0.333]`.
  70. 3. `"$PY" cut.py render --run "<run folder>"` builds every clip: face cam on top (40%), gameplay underneath, the hook in a yellow box for the first 3 seconds, word-by-word captions from a fresh Whisper pass on the clip itself, audio normalized to Shorts loudness, and saves to `<run>/clips/`. Every finished clip is also copied to `WORK/videos/` with a serial number that continues across all runs (`001_<hook>.mp4`, `002_...`); a re-rendered clip keeps its number. Point the user at `WORK/videos/` as the place their clips are. Then it checks each clip for length, a black face cam area, black gameplay and missing captions.
  71. 4. If a check fails, or the user wants a change, edit `<run>/layout.json` or `<run>/picks.json` (the copy inside the run folder, not `WORK/picks.json`) and re-render just that clip: `"$PY" cut.py render --run "<run>" --only 03`. A new hook, title or caption only needs the re-render. New times need `"$PY" cut.py download --run "<run>" --only 03` first, then a fresh look at its frames, then the re-render. If a clip prints `!! render failed`, the other clips are still fine; fix that one clip's layout or times and re-render it alone.
  72. 5. Look at a frame from the start and the middle of each finished clip yourself (`~/.clipper/bin/ffmpeg -ss 2 -i clip.mp4 -frames:v 1 check.jpg`, then again at the midpoint) before you say it's done. If the face is cut off, the crop is wrong or the captions read wrong, fix it and re-render.
  73. ## 5. Post
  74. `"$PY" gallery.py --run "<run folder>"` opens one page with every clip (best first), its title, caption and hashtags, copy buttons and a link to the exact second in the source stream.
  75. Tell the user, every time, in one or two lines:
  76. - Watch each clip before posting and cut anything that starts slow.
  77. - Add something of their own (their hook, a reaction, a take). Channels of raw reposted clips get turned down for monetization; "the clips are the reach, you are the channel".
  78. - Credit the streamer or creator, follow the clipping campaign's rules, and don't post the same clip twice on one account.
  79. There is no auto-upload in this skill. Posting stays with the user so nothing goes out unreviewed.
  80. ## 6. Look at what's working (weekly)
  81. `"$PY" learn.py --channel https://www.youtube.com/@<their handle>`
  82. Reads the public views on their Shorts tab, matches each Short to the clips in `WORK/log.csv` by title, and reports average views by moment type and by streamer, plus the top 5 and bottom 5. It only matches Shorts whose title still contains the clip's title or hook, so tell users to keep that.
  83. Then update `WORK/preferences.md` with 2 to 5 concrete rules from the data ("freak-outs do 2.1x my average, prefer them", "Kai clips beat xQc clips 3 to 1", "hooks that name the streamer beat hooks that don't"), and show the user what you changed. Section 3 reads this file every time, so the picks get better each week. Don't invent patterns from fewer than 10 matched clips; say there isn't enough data yet.
  84. ## Troubleshooting
  85. - **"This stream is still live"**: analyze.py stops on live streams and on YouTube recordings that are still processing. Tell the user to come back once the stream has ended and the recording is up, or pick a different stream.
  86. - **"Sign in to confirm you're not a bot" or 403 from YouTube:** set the browser to borrow cookies from (`export CLIPPER_COOKIES_BROWSER=chrome`, or safari/firefox) and run again. Also run setup again to update yt-dlp, which fixes most YouTube breakages.
  87. - **No chat speed:** only YouTube VODs of live streams have chat replay. Loudness and most-replayed still work.
  88. - **Most-replayed missing:** new or low-view videos don't have the graph yet. Loudness carries it.
  89. - **A clip has no captions:** there was no clear speech (a pure scream). That's fine; the hook still shows.
  90. - **Captions look wrong:** captions already use the `small` Whisper model. If a line is still wrong (loud music, people talking over each other), re-render that clip with `--only <key> --model medium` (slower, downloads once), or trim the clip around it.
  91. - **Very slow transcribing:** use `--model base` or `--top 15`.
  92. - Everything is in `WORK`. A stream's audio can be deleted after its clips are made.
  93. ## What not to do
  94. - Never clip or post anything the user hasn't seen; never claim a clip is good without looking at it.
  95. - Never pick from a stream the user doesn't have the right to clip if they tell you a campaign's rules forbid it.
  96. - Never make up numbers about how a clip will perform.
README.md · README · 133 lines
  1. # Stream Clipper
  2. A free [Claude Code](https://claude.com/claude-code) skill that turns any long stream, podcast or video into ready-to-post vertical clips. It was built for GTA 6 streams, and it works the same on any game, IRL and Just Chatting streams, podcasts and interviews.
  3. Give Claude a link, or ask it to find today's biggest streams on any topic. It listens to the whole stream, picks the best moments with a harsh filter, and cuts 9:16 clips with the face cam on top and the main video underneath (or one clean crop when there's no face cam, like most podcasts), a hook and word-by-word captions. Then it opens one page where you watch, copy the caption and post.
  4. In testing, an hour-long stream turned into 3 ready-to-post clips in about 13 minutes on a MacBook.
  5. ![The pipeline: find, listen, pick, cut, post](docs/pipeline.png)
  6. Based on [Clipper](https://github.com/raycfu/clipper) by Ray Fu ([@raycfu](https://www.instagram.com/raycfu)), MIT licensed. This version adds sound-event detection and a numbered videos folder.
  7. ## What's new in this version
  8. - **It hears what's happening, not just how loud it is.** A free sound classifier ([YAMNet](https://huggingface.co/andrelgomes/yamnet-onnx), runs on your computer, ~16 MB) scores every second of every stream for action (gunfire, explosions, crashes, skids, sirens, helicopters) and reactions (screams, shouts, laughter). Plain talking with no action, like a game cutscene, scores lower, so cutscenes stop beating real moments. An 11-hour stream takes about 2 extra minutes.
  9. - **Reaction and game words.** Every window gets a bonus for words like "oh my god", "bro what", "no no no". When the title names a game, that game's words count too (GTA: cops, wanted, wasted, busted, heist, chase, crash...). Add your own game in `scripts/profiles.py`.
  10. - **Chat replay retry.** Long chat replays that cut off halfway get downloaded again.
  11. - **40 candidates** instead of 25, listed strongest first, each with what the sound classifier heard.
  12. - **All clips in one folder, numbered.** Every finished clip is also copied to `~/Clipper/videos/` as `001_<hook>.mp4`, `002_...`, continuing across every stream you clip. A re-rendered clip keeps its number.
  13. ## What you need
  14. - A Mac or Linux computer. On Windows, run Claude Code inside [WSL](https://learn.microsoft.com/windows/wsl/install).
  15. - [Claude Code](https://claude.com/claude-code), which comes with the Claude Pro plan ($20/month).
  16. - About 3 GB of free space for the tools and speech models, plus room for the streams you clip (a few hundred MB each).
  17. Everything else is free and installs itself the first time: yt-dlp, ffmpeg, Whisper (speech to text that runs on your own computer), deno and a caption font. Nothing is installed system-wide. It all lives in `~/.clipper`.
  18. ## Install
  19. ```bash
  20. git clone https://github.com/twahanur/stream-clipper ~/.claude/skills/clipper
  21. ```
  22. No git? Click **Code → Download ZIP** above, unzip it, rename the folder to `clipper` and move it into `~/.claude/skills/`. On a Mac that folder is hidden: in Finder press `Cmd + Shift + G`, paste `~/.claude/skills/` and press Enter.
  23. Then open Claude Code (in the terminal, or the Code tab of the Claude desktop app) and say:
  24. > set up the clipper
  25. Setup takes about 5 minutes the first time and ends with `READY`. Allow the commands when Claude asks.
  26. ## Use it
  27. Say one of these to Claude:
  28. | Say | What happens |
  29. |---|---|
  30. | `clip this stream: <link>` | Clips a finished YouTube, Twitch or Kick stream, a podcast or any long video |
  31. | `find me today's biggest GTA 6 streams and clip them` | Finds the most viewed streams first, then clips them. Swap GTA 6 for any game, streamer or topic |
  32. | `find me this week's biggest podcast episodes and clip them` | Same thing for podcasts |
  33. | `make the hook for clip 2 shorter` | Changes one clip and re-renders only that one |
  34. | `clip 3 starts slow, start it 4 seconds later` | Re-cuts one clip |
  35. | `look at what's working` | Weekly report from your posted Shorts' views (give it your channel link) |
  36. Claude shows you a table of the moments it kept and the ones it cut, with a reason for each, then renders the clips and opens the gallery page. Clips are saved in `~/Clipper/videos/` (numbered 001, 002, ...) and in `~/Clipper/runs/<date>/clips/`.
  37. ## How it works
  38. 1. **Find**: searches YouTube for the most viewed long streams or videos on your topic. Streams that are still live are skipped until the recording is up.
  39. 2. **Listen**: downloads only the audio, then builds a hype curve from how loud the streamer gets, what the sound classifier hears (action and reaction sounds), YouTube's "most replayed" graph and, on YouTube stream recordings with chat replay, how fast chat is moving. It transcribes the 40 hottest spots with Whisper on your computer and scores them for reaction and game words.
  40. 3. **Pick**: Claude scores every moment with three questions. Would a stranger stop scrolling in two seconds? Does it make sense with zero context? Does it pay off inside the clip? One no and it's cut.
  41. 4. **Cut**: downloads only the minutes each clip needs and builds a 1080x1920 video: face cam on top and the main video underneath (or one full-frame crop when there's no face cam), a hook for the first 3 seconds, word-by-word captions and audio at Shorts loudness. Claude checks three frames of every clip and trims around any layout switch in the stream, and every clip is checked for length, a black face cam and missing captions.
  42. 5. **Post**: one page with every clip, a title, caption and hashtags with copy buttons, and a link to the exact second in the original stream so you can credit it. Nothing is ever posted for you.
  43. Every week, `look at what's working` reads the public views on your Shorts, matches them to the clips the skill made, and updates `~/Clipper/preferences.md`, so the picks lean toward what your audience actually watches. It needs about 10 posted Shorts, and it only matches Shorts whose title still contains the clip's title or hook.
  44. ## Before you post
  45. - Watch every clip first.
  46. - Add something of your own: your hook, your take or a reaction. Channels of raw reposted clips get turned down for monetization.
  47. - Credit the streamer or creator, follow the clipping campaign's rules, and don't post the same clip twice on one account.
  48. - Only clip streams you're allowed to clip. You're responsible for what you post.
  49. ## Update
  50. ```bash
  51. cd ~/.claude/skills/clipper && git pull
  52. ```
  53. Then say "set up the clipper" again. It takes a few seconds and also updates the downloader, which fixes most YouTube download errors.
  54. ## Uninstall
  55. ```bash
  56. rm -rf ~/.claude/skills/clipper ~/.clipper
  57. ```
  58. Your clips stay in `~/Clipper` until you delete that folder too.
  59. ## Troubleshooting
  60. - **"Sign in to confirm you're not a bot" or a 403 error**: say "set up the clipper" again to update the downloader. If it still fails, run `export CLIPPER_COOKIES_BROWSER=chrome` (or `safari` / `firefox`, whichever browser you're signed into YouTube with) and try again.
  61. - **"This video is age-restricted"**: same fix, `export CLIPPER_COOKIES_BROWSER=chrome`.
  62. - **"This stream is still live"**: wait until the stream ends and the recording is up.
  63. - **No chat signal**: chat speed only works on YouTube recordings of live streams that kept chat replay on. Twitch and Kick use loudness only, which still works well.
  64. - **A caption is wrong**: ask Claude to re-render that clip with the `medium` model.
  65. - **Transcribing is slow**: ask Claude to use the `base` model or fewer candidates (`--top 15`).
  66. - **Something else**: [open an issue](https://github.com/twahanur/stream-clipper/issues) with the last lines Claude showed you.
  67. ## FAQ
  68. **Does it cost anything besides Claude Pro?** No. Every tool it uses is free and runs on your computer.
  69. **Does it work for games other than GTA 6?** Yes. Nothing in it is tied to one game. Paste any stream link, or ask it to find streams for any game or streamer.
  70. **Podcasts and interviews?** Yes. With no face cam it cuts one clean vertical crop on the person speaking, and it picks moments by what's said (stories, advice, hot takes), not just by how loud it gets.
  71. **I installed the old GTA 6 Clipper. Do I need to do anything?** Delete `~/.claude/skills/gta6-clipper`, install again with the command above and say "set up the clipper". Your old clips stay in `~/GTA6 Clipper`.
  72. **Does it upload anything?** No. It never posts, and your clips never leave your computer until you post them.
  73. **Windows?** Use WSL. Native Windows isn't supported, because the setup script needs bash.
  74. ## Files
  75. ```
  76. SKILL.md the instructions Claude follows
  77. scripts/setup.sh one-time install into ~/.clipper
  78. scripts/find_streams.py find the biggest streams or videos on any topic
  79. scripts/analyze.py audio, hype curve, transcripts of the hot spots
  80. scripts/sound_events.py sound classifier: gunfire, explosions, crashes, screams, laughter
  81. scripts/profiles.py reaction words and per-game word lists (GTA)
  82. scripts/tests/ quick tests (python scripts/tests/test_*.py)
  83. scripts/cut.py download only the needed minutes, render vertical clips
  84. scripts/gallery.py the page you post from
  85. scripts/learn.py weekly "look at what's working" report
  86. scripts/common.py shared helpers
  87. ```
  88. ## License
  89. MIT. See [LICENSE](LICENSE). Original Clipper by Ray Fu; sound events, word lists and the videos folder added in this version. YAMNet is Apache-2.0 (Google).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.