Claude Skill

conference-talk

Conference talk editor for Claude: turns a keynote, panel, meetup or webinar recording (speaker camera plus slides) into a clean 16:9 session video with a title card, speaker lower third, slide inserts, trimmed dead air and applause, accessible captions and chapters, plus a 60-90

LLM Mart · 0 points · 0 views 0 listing impressions 0 install-command copies
Download viprasol-tech-claude-video-editing-skills-plugins_video-studio_skills_conference-talk-0293497.zip · 4 KB

Install

skills CLI npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/conference-talk
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
Git git clone https://github.com/Viprasol-Tech/claude-video-editing-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole viprasol-tech/claude-video-editing-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Conference Talk

The talk as the audience should have seen it: the speaker when they matter, the slide when it matters, and nothing lost.

  • Best for: keynotes, conference sessions, meetups, panels, webinars, lectures, TEDx-style talks
  • Inputs: speaker camera (or the event's program feed), the slide recording or slide images, speaker name/title, event logo
  • Default skin: chapter-pro · alternatives minimal-mono, news-desk, frame-studio
  • Formats: 16:9 full session; 16:9 highlight; 9:16 quote clips
  • Durations: session 10-60 min as given; highlight 60-90 s; quote clips 20-45 s
  • Genre preset: --genre conference-talk

Before you start

  • Needs video-studio-core in the same skills folder (CORE); SKILLS = the skills folder.
  • First run: python CORE/scripts/check_env.py.
  • Ask for: exact speaker name and title, session title, event name and date, whether Q&A stays in, the slides (PDF exported to PNG, or a screen recording), and the event's permission to publish.
  • Audience members asking questions in Q&A did not necessarily agree to be published: cut or blur them unless the event has consent, and keep only what the speaker repeats.
  • Slides may contain third-party images or confidential numbers; ask before publishing every slide.

How a pro cuts this

Structure (session, % of length)

beat % what
title card 2 session title, speaker, event, date (3-5 s)
talk 90 the talk in order; trim walk-on, mic fiddling, "can you hear me", long applause
Q&A 6 keep the best 1-3 questions, repeat the question as a caption if the mic missed it
end card 2 event logo, links, next event

Highlight (60-90 s): the strongest 3-5 lines in talk order, each a complete thought, cold open on the best line.

Pacing. This is long-form for a patient audience: ASL 6-20 s, a visual change at least every 20 s (slide insert, punch-in, wide). Do not tighten pauses like a reel: the speaker's timing is part of the talk. Max gap 1.2 s, fillers kept natural (only um/uh go), no retake removal (a talk has no retakes).

Cut rules.

  • Cut to the slide when the speaker says "as you can see", reads a number or points at the screen; hold the slide at least 4-6 s (people read it), longer for dense slides; go back to the speaker for stories, jokes and emotion.
  • Slide + speaker picture-in-picture for long slide sections so the face stays present.
  • Trim applause to 1-2 s after the opening and the ending; never cut a laugh off.
  • Cover the few autocut joins with a slide insert or a punch-in (>= 1.2x) so no jump cut shows.

Transition grammar. Straight cuts between speaker and slide (a 0.3 s dissolve at most), dip to black between sessions or before Q&A. Nothing else.

Music and sound. Music on the title and end cards only (bed -30 dB). Under the talk: none. Clean the room: light denoise, broadcast EQ, de-ess; level the speaker; keep audience laughter. SFX none.

Colour. Accurate, not stylised: correct the white balance of stage light, cool-corporate or documentary-natural at 0.4-0.5. Match the slide recording's brightness to the camera so cuts don't flash.

Captions. Accessible subtitles, sentence case, two lines, <= 15 cps ("preset": "clean"), plus sidecar SRT/VTT for the platform. Speaker name as a lower third at first appearance.

What the client judges. Slides readable, the speaker looks good, the sound is clear, names right, and it is on the website quickly.

Common mistakes. Slides too small or on screen too briefly; the speaker cut out while telling a story; reel-style tightening; loud music under speech; missing lower third; a wrong job title; Q&A questions inaudible.

Make the edit

# speaker camera + slide recording (available to elements as clip id "main")
python CORE/scripts/run.py talk.mp4 --style SKILLS/chapter-pro --genre conference-talk --format 16:9 --clip slides.mp4 --title "Three Mistakes" --model base --work ct --stop-after plan
# where do the slides change? (shot list of the slide recording; thumbs to look at)
python CORE/scripts/scenes.py slides.mp4 --work ct --threshold 0.15 --thumbs --out ct/slides_scenes.json
python CORE/scripts/story.py outline --work ct --genre conference-talk    # sentences s1..sN, quotable lines for the highlight
python CORE/scripts/render.py ct/plan.json --frames 1,4,8,13,16
python CORE/scripts/render.py ct/plan.json
python CORE/scripts/finish.py ct/plan.json --platform youtube

The genre preset keeps pauses up to 1.2 s (a speaker's timing); pass --max-gap only to change that. For names and jargon add --prompt "Priya Rao, Kubernetes". Long talks: --model small (default) is more accurate.

Separate camera + program feed: if you have a wide and a close-up with the same audio, multicam.py will not help (it switches by loudest mic). Cut them by hand in a timeline spec, or use the program feed as the single input.

Highlight: pick sentences from ct/outline.md, then python CORE/scripts/story.py reorder --work ct --order "s12,s3-s4,s20" --genre conference-talk --format 16:9 and python CORE/scripts/run.py - --timeline ct/story_timeline.json --style SKILLS/chapter-pro --genre conference-talk --format 16:9 --model base --work ch --stop-after plan. Read ct/story_flags.md: no partial-sentence splices in a talk (it is the speaker's real words).

Quote clips (9:16): use clip-miner on the finished talk, or story.py reorder ... --format 9:16 per quote.

Refine the plan

  • Captions: set the captions element to {"type": "captions", "preset": "clean", "maxWords": 8}; fix names in words.
  • Remove auto extras that don't fit a talk: auto counter elements (check the value and label match what was said), chip "Subscribe", list cards built from the transcript.
  • Title card: {"type": "titlecard", "kicker": "VIDEO STUDIO CONF 2026", "title": "Three Mistakes", "subtitle": "Priya Rao, Founder", "start": 0, "end": 3}.
  • Lower third: {"type": "lowerthird", "name": "Priya Rao", "role": "Founder, Studio Notes", "start": 3.2, "end": 8, "in": "maskUp", "out": "fade"}.
  • Full-screen slide insert (speaker audio continues): "cutaways": [{"src": "C:/talks/slides.mp4", "start": 6.0, "end": 10.5, "srcStart": 42.0}] with srcStart = the slide's time in the slide recording (from slides_scenes.json). Check the insert lands on the right slide in the --frames stills.
  • Picture-in-picture slide: {"type": "clip", "clip": "main", "clipOffset": 42.0, "start": 12, "end": 16.5, "x": 1530, "y": 640, "w": 640, "h": 360, "in": "fade", "out": "fade"}. Keep it off the face (look at the sheet). Slide images instead of a recording: {"type": "image", "image": "slide07.png", ...}.
  • Chapters: "chapters": [{"start": 0, "title": "Intro"}, {"start": 95, "title": "Mistake 1"}, ...] (first at 0, >= 3, each >= 10 s).
  • Colour and sound: "color": {"look": "cool-corporate", "intensity": 0.5}, "finish": {"sharpen": 0.3}, "audio": {"platform": "youtube", "voice": {"denoise": "light", "eq": "broadcast", "deess": true}}. The plan's audio.platform wins over finish.py --platform, so set the target in one place.
  • Intro/outro music only: there is no per-section music in the plan. Build the title card + talk + end card as a timeline spec with the music on the card clips, or leave music out.
  • python CORE/scripts/editor.py review ct/plan.json --genre conference-talk and fix its high items.

Deliver

python CORE/scripts/deliver.py ct/plan.json --to youtube,podcast-mp3 --out-dir ct/deliver --burn no
python CORE/scripts/qc.py ct/deliver/final_youtube.mp4 --plan ct/plan.json --platform youtube --strict
python CORE/scripts/qc.py ct/deliver/final_podcast-mp3.mp3 --platform podcast

Upload captions.srt/.vtt as sidecars (YouTube, LMS, Vimeo) and paste chapters.txt into the description. For an event site or LMS that wants -16 LUFS use podcast-video instead of youtube. Name files event2026_priya-rao_three-mistakes_v001_16x9.mp4.

Check before delivering

  • Slides readable at 1080p on a phone; slide inserts held long enough to read; no slide covered by a lower third.
  • Speaker name, title and session title spelled exactly as the event program.
  • No jump cut visible; applause trimmed but laughs intact; the opening line is not cut.
  • Captions match the audio (names, numbers, jargon); qc.py passes caption timing and loudness.
  • Q&A voices audible or the question captioned; no audience member shown without consent.

Example requests

  • "Edit my keynote recording, put my slides in when I'm talking about them, captions on."
  • "Clean up this 40-minute webinar: cut the setup, keep Q&A, add chapters."
  • "Make a 90-second highlight of my talk for LinkedIn."
Files (claude-video-editing-skills)
  • SKILL.md 9.3 KB
    ---
    name: conference-talk
    description: "Conference talk editor for Claude: turns a keynote, panel, meetup or webinar recording (speaker camera plus slides) into a clean 16:9 session video with a title card, speaker lower third, slide inserts, trimmed dead air and applause, accessible captions and chapters, plus a 60-90 s highlight and vertical quote clips. Use when the user says 'edit my talk', 'clean up this keynote recording', 'conference session video', 'add my slides to the talk', 'webinar recording edit', 'make a highlight from my speech', 'TEDx style edit'. Runs on video-studio-core (transcribe, gentle autocut, scenes, colour, deliver, QC)."
    license: MIT
    ---
    
    # Conference Talk
    
    The talk as the audience should have seen it: the speaker when they matter, the slide when it matters, and nothing lost.
    
    - **Best for:** keynotes, conference sessions, meetups, panels, webinars, lectures, TEDx-style talks
    - **Inputs:** speaker camera (or the event's program feed), the slide recording or slide images, speaker name/title, event logo
    - **Default skin:** `chapter-pro` · alternatives `minimal-mono`, `news-desk`, `frame-studio`
    - **Formats:** 16:9 full session; 16:9 highlight; 9:16 quote clips
    - **Durations:** session 10-60 min as given; highlight 60-90 s; quote clips 20-45 s
    - **Genre preset:** `--genre conference-talk`
    
    ## Before you start
    
    - Needs **video-studio-core** in the same skills folder (`CORE`); `SKILLS` = the skills folder.
    - First run: `python CORE/scripts/check_env.py`.
    - Ask for: exact speaker name and title, session title, event name and date, whether Q&A stays in, the slides (PDF
      exported to PNG, or a screen recording), and the event's permission to publish.
    - Audience members asking questions in Q&A did not necessarily agree to be published: cut or blur them unless the event
      has consent, and keep only what the speaker repeats.
    - Slides may contain third-party images or confidential numbers; ask before publishing every slide.
    
    ## How a pro cuts this
    
    **Structure (session, % of length)**
    
    | beat | % | what |
    | --- | --- | --- |
    | title card | 2 | session title, speaker, event, date (3-5 s) |
    | talk | 90 | the talk in order; trim walk-on, mic fiddling, "can you hear me", long applause |
    | Q&A | 6 | keep the best 1-3 questions, repeat the question as a caption if the mic missed it |
    | end card | 2 | event logo, links, next event |
    
    Highlight (60-90 s): the strongest 3-5 lines in talk order, each a complete thought, cold open on the best line.
    
    **Pacing.** This is long-form for a patient audience: ASL 6-20 s, a visual change at least every 20 s (slide insert,
    punch-in, wide). Do not tighten pauses like a reel: the speaker's timing is part of the talk. Max gap 1.2 s, fillers kept
    natural (only um/uh go), no retake removal (a talk has no retakes).
    
    **Cut rules.**
    - Cut to the slide when the speaker says "as you can see", reads a number or points at the screen; hold the slide at
      least 4-6 s (people read it), longer for dense slides; go back to the speaker for stories, jokes and emotion.
    - Slide + speaker picture-in-picture for long slide sections so the face stays present.
    - Trim applause to 1-2 s after the opening and the ending; never cut a laugh off.
    - Cover the few autocut joins with a slide insert or a punch-in (>= 1.2x) so no jump cut shows.
    
    **Transition grammar.** Straight cuts between speaker and slide (a 0.3 s dissolve at most), dip to black between
    sessions or before Q&A. Nothing else.
    
    **Music and sound.** Music on the title and end cards only (bed -30 dB). Under the talk: none. Clean the room: light
    denoise, `broadcast` EQ, de-ess; level the speaker; keep audience laughter. SFX none.
    
    **Colour.** Accurate, not stylised: correct the white balance of stage light, `cool-corporate` or
    `documentary-natural` at 0.4-0.5. Match the slide recording's brightness to the camera so cuts don't flash.
    
    **Captions.** Accessible subtitles, sentence case, two lines, <= 15 cps (`"preset": "clean"`), plus sidecar SRT/VTT for
    the platform. Speaker name as a lower third at first appearance.
    
    **What the client judges.** Slides readable, the speaker looks good, the sound is clear, names right, and it is on
    the website quickly.
    
    **Common mistakes.** Slides too small or on screen too briefly; the speaker cut out while telling a story; reel-style
    tightening; loud music under speech; missing lower third; a wrong job title; Q&A questions inaudible.
    
    ## Make the edit
    
    ```bash
    # speaker camera + slide recording (available to elements as clip id "main")
    python CORE/scripts/run.py talk.mp4 --style SKILLS/chapter-pro --genre conference-talk --format 16:9 --clip slides.mp4 --title "Three Mistakes" --model base --work ct --stop-after plan
    # where do the slides change? (shot list of the slide recording; thumbs to look at)
    python CORE/scripts/scenes.py slides.mp4 --work ct --threshold 0.15 --thumbs --out ct/slides_scenes.json
    python CORE/scripts/story.py outline --work ct --genre conference-talk    # sentences s1..sN, quotable lines for the highlight
    python CORE/scripts/render.py ct/plan.json --frames 1,4,8,13,16
    python CORE/scripts/render.py ct/plan.json
    python CORE/scripts/finish.py ct/plan.json --platform youtube
    ```
    
    The genre preset keeps pauses up to 1.2 s (a speaker's timing); pass `--max-gap` only to change that.
    For names and jargon add `--prompt "Priya Rao, Kubernetes"`. Long talks: `--model small` (default) is more accurate.
    
    **Separate camera + program feed:** if you have a wide and a close-up with the same audio, `multicam.py` will not help
    (it switches by loudest mic). Cut them by hand in a timeline spec, or use the program feed as the single input.
    
    **Highlight:** pick sentences from `ct/outline.md`, then
    `python CORE/scripts/story.py reorder --work ct --order "s12,s3-s4,s20" --genre conference-talk --format 16:9` and
    `python CORE/scripts/run.py - --timeline ct/story_timeline.json --style SKILLS/chapter-pro --genre conference-talk --format 16:9 --model base --work ch --stop-after plan`.
    Read `ct/story_flags.md`: no partial-sentence splices in a talk (it is the speaker's real words).
    
    **Quote clips (9:16):** use clip-miner on the finished talk, or `story.py reorder ... --format 9:16` per quote.
    
    ## Refine the plan
    
    - **Captions:** set the captions element to `{"type": "captions", "preset": "clean", "maxWords": 8}`; fix names in `words`.
    - **Remove auto extras** that don't fit a talk: auto `counter` elements (check the value and label
      match what was said), `chip` "Subscribe", list `card`s built from the transcript.
    - **Title card:** `{"type": "titlecard", "kicker": "VIDEO STUDIO CONF 2026", "title": "Three Mistakes", "subtitle": "Priya Rao, Founder", "start": 0, "end": 3}`.
    - **Lower third:** `{"type": "lowerthird", "name": "Priya Rao", "role": "Founder, Studio Notes", "start": 3.2, "end": 8, "in": "maskUp", "out": "fade"}`.
    - **Full-screen slide insert** (speaker audio continues):
      `"cutaways": [{"src": "C:/talks/slides.mp4", "start": 6.0, "end": 10.5, "srcStart": 42.0}]` with `srcStart` = the
      slide's time in the slide recording (from `slides_scenes.json`). Check the insert lands on the right slide in the `--frames` stills.
    - **Picture-in-picture slide:** `{"type": "clip", "clip": "main", "clipOffset": 42.0, "start": 12, "end": 16.5, "x": 1530, "y": 640, "w": 640, "h": 360, "in": "fade", "out": "fade"}`.
      Keep it off the face (look at the sheet). Slide images instead of a recording: `{"type": "image", "image": "slide07.png", ...}`.
    - **Chapters:** `"chapters": [{"start": 0, "title": "Intro"}, {"start": 95, "title": "Mistake 1"}, ...]` (first at 0, >= 3, each >= 10 s).
    - **Colour and sound:** `"color": {"look": "cool-corporate", "intensity": 0.5}`, `"finish": {"sharpen": 0.3}`,
      `"audio": {"platform": "youtube", "voice": {"denoise": "light", "eq": "broadcast", "deess": true}}`.
      The plan's `audio.platform` wins over `finish.py --platform`, so set the target in one place.
    - Intro/outro music only: there is no per-section music in the plan. Build the title card + talk + end card as a
      timeline spec with the music on the card clips, or leave music out.
    - `python CORE/scripts/editor.py review ct/plan.json --genre conference-talk` and fix its high items.
    
    ## Deliver
    
    ```bash
    python CORE/scripts/deliver.py ct/plan.json --to youtube,podcast-mp3 --out-dir ct/deliver --burn no
    python CORE/scripts/qc.py ct/deliver/final_youtube.mp4 --plan ct/plan.json --platform youtube --strict
    python CORE/scripts/qc.py ct/deliver/final_podcast-mp3.mp3 --platform podcast
    ```
    
    Upload `captions.srt`/`.vtt` as sidecars (YouTube, LMS, Vimeo) and paste `chapters.txt` into the description. For an
    event site or LMS that wants -16 LUFS use `podcast-video` instead of `youtube`. Name files
    `event2026_priya-rao_three-mistakes_v001_16x9.mp4`.
    
    ## Check before delivering
    
    - Slides readable at 1080p on a phone; slide inserts held long enough to read; no slide covered by a lower third.
    - Speaker name, title and session title spelled exactly as the event program.
    - No jump cut visible; applause trimmed but laughs intact; the opening line is not cut.
    - Captions match the audio (names, numbers, jargon); `qc.py` passes caption timing and loudness.
    - Q&A voices audible or the question captioned; no audience member shown without consent.
    
    ## Example requests
    
    - "Edit my keynote recording, put my slides in when I'm talking about them, captions on."
    - "Clean up this 40-minute webinar: cut the setup, keep Q&A, add chapters."
    - "Make a 90-second highlight of my talk for LinkedIn."
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.