reaction-duet
Reaction / duet / stitch editor: combines a source clip with the creator's facecam reaction in a picture-in-picture, split-screen or stitch layout, keeps the two in sync, ducks the source audio under the reactor's voice, freezes or zooms on the moment being reacted to, and delive
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/reaction-duet
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
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
Reaction Duet
Source on one side, you on the other, perfectly in sync, the source politely ducking whenever you talk.
- Best for: reactions, commentary, duets, stitches, "expert reacts", breakdowns of someone else's clip
- Inputs: the source clip (
source.mp4, yours to use or used under fair use with commentary), the facecam recording made while watching it, the source offset (where in the source your recording starts) - Default skin:
split-proof(facecam bottom half) · alternativeshalo-cut(PiP),bold-bounce,before-after - Formats: 9:16 1080x1920 with captions;
captions.srt - Duration target: 15-60 s Shorts, up to 3 min; genre preset
reaction-duet
Before you start
- Needs video-studio-core in the same skills folder.
CORE=video-studio-core,SKILLS= the folder holding both. - First time:
python CORE/scripts/check_env.py. Preset:python CORE/scripts/editor.py genre reaction-duet. - Sync first. Record the facecam while the source plays, clap or say "go" on its first frame, and note the offset. Work out
OFFSET= source second that plays at facecam second 0. - Ethics and rights: credit the creator on screen for the whole clip, add real commentary (not just watching), never fake a reaction, use short excerpts. Ask the user if they have the creator's permission when the clip is long.
How a pro cuts this
| beat | % | what happens |
|---|---|---|
| hook | 0-10 | the most surprising moment of the source + your strongest reaction, in the first 1.5 s |
| context | 10-20 | who / what this is, one line, credit on screen |
| source + reactions | 20-80 | play the source; pause (freeze) or zoom where you talk over it |
| verdict | 80-95 | your take: the value you add |
| CTA | last 2 s | follow / part 2 |
- Sync is sacred: do NOT run autocut on a reaction (
--no-cut); every removed pause would desync your face from the source. Cut only whole sections with--cutranges and move the source the same amount (a new clip element with its ownclipOffset). - Pacing: ASL 2-4 s; a change every 3 s: zoom on the source detail, freeze-frame + callout, a punch-in on your face.
- Audio: source audio plays as a bed ducked ~10 dB whenever you speak (the plan
musicblock does this); your voice leads. - Transitions: cuts only; flash on the reveal moment at most once.
- Colour: correct the facecam only (
coloraffects the speaker video); leave the source as is. - Captions: your words only (the source can carry its own), preset
boxedorhormozi, in the band between the two panes. - Client judges: is the reaction genuine, does the commentary add value, hold rate.
- Common mistakes: drifting sync, source louder than the reactor, no credit, reacting with nothing to say, captions over the source's own subtitles.
Make the edit
python CORE/scripts/run.py facecam.mp4 --style SKILLS/split-proof --genre reaction-duet --no-cut \
--clip source.mp4 --format 9:16 --work react --stop-after plan
python CORE/scripts/render.py react/plan.json --frames 1,4,8,14
--clip source.mp4 makes the source available to elements as clip id main. Then wire the layout and the audio in the plan (next section), review and render:
python CORE/scripts/editor.py review react/plan.json --genre reaction-duet --fix
python CORE/scripts/render.py react/plan.json --frames 1,4,8,14 # LOOK: source visible, face visible, credit
python CORE/scripts/render.py react/plan.json
python CORE/scripts/finish.py react/plan.json --platform tiktok
Refine the plan
- Source pane (split, top half):
{"type": "clip", "clip": "main", "clipOffset": OFFSET, "start": 0, "end": DUR, "x": 540, "y": 480, "w": 1080, "h": 960, "in": "none"}. The facecam already fills the bottom half insplit-proof. - PiP instead: use
halo-cut(face full frame) and a smaller source:"w": 620, "h": 350, "x": 540, "y": 520, "className": "glass"; or swap: source full (w/h= frame) and set"speaker": {"mode": "panel", "rect": [700, 1300, 330, 440], "radius": 40}for the face. - Source audio, ducked:
"music": {"src": "../source.mp4", "start": OFFSET, "volume": 0.5, "duck": true, "duckDepth": 10, "fadeOut": 0.5}.startis the source second at video 0, so it matchesclipOffset. - Freeze on a moment: end the playing clip element at T and add an
imageelement (a still exported withffmpeg -ss <src time> -i source.mp4 -frames:v 1 freeze.png) from T to T+2, then a new clip element withclipOffset= OFFSET + T + 2 (the source jumps past what was frozen; adjustmusic.startthe same way only if you split the audio too). - Callouts:
{"type": "callout", "point": [700, 420], "x": 300, "y": 180, "title": "watch his hand"},scribblecircles on the detail. - Credit:
{"type": "chip", "text": "Source: @creator", "start": 0, "end": DUR, "x": 60, "y": 120, "anchor": "top-left"}. - Captions:
{"type": "captions", "preset": "boxed", "y": 1000, "dodgeFace": false}- place between panes.
Deliver
python CORE/scripts/deliver.py react/plan.json --to tiktok,reels,shorts --out-dir react/deliver
python CORE/scripts/qc.py react/deliver/final_tiktok.mp4 --plan react/plan.json --platform tiktok
- Put the original link in
post.md's description. Platform-native duet/stitch features keep attribution automatically; this edit is for re-uploads with commentary.
Check before delivering
- Lips and the source stay in sync at the start, middle and end (compare 3 stills).
- Reactor always intelligible; source audible in gaps; no clipping (
qc.py-14 LUFS / -1 dBTP). - Credit on screen the whole time; commentary present in every section.
- Captions don't cover the source's own text or the face.
- Tell the user: layout, offset used, length, file paths, and the rights caveat.
Example requests
- "Make a reaction video: this clip on top, my face below, duck the clip when I talk."
- "Stitch-style: first 3 seconds of their video, then my answer."
- "PiP my facecam over this trailer, freeze when I explain the shot."
Files (claude-video-editing-skills)
-
SKILL.md 6.6 KB
--- name: reaction-duet description: "Reaction / duet / stitch editor: combines a source clip with the creator's facecam reaction in a picture-in-picture, split-screen or stitch layout, keeps the two in sync, ducks the source audio under the reactor's voice, freezes or zooms on the moment being reacted to, and delivers a 9:16 short with captions and an on-screen credit. Use when the user says: reaction video, react to this clip, duet, stitch, split screen with my face, commentary on a video, PiP facecam over a clip. Commentary must add something (fair use) and the source must be credited. Default skin split-proof." license: MIT --- # Reaction Duet Source on one side, you on the other, perfectly in sync, the source politely ducking whenever you talk. - **Best for:** reactions, commentary, duets, stitches, "expert reacts", breakdowns of someone else's clip - **Inputs:** the source clip (`source.mp4`, yours to use or used under fair use with commentary), the facecam recording made while watching it, the source offset (where in the source your recording starts) - **Default skin:** `split-proof` (facecam bottom half) · alternatives `halo-cut` (PiP), `bold-bounce`, `before-after` - **Formats:** 9:16 1080x1920 with captions; `captions.srt` - **Duration target:** 15-60 s Shorts, up to 3 min; genre preset `reaction-duet` ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = `video-studio-core`, `SKILLS` = the folder holding both. - First time: `python CORE/scripts/check_env.py`. Preset: `python CORE/scripts/editor.py genre reaction-duet`. - **Sync first.** Record the facecam while the source plays, clap or say "go" on its first frame, and note the offset. Work out `OFFSET` = source second that plays at facecam second 0. - **Ethics and rights:** credit the creator on screen for the whole clip, add real commentary (not just watching), never fake a reaction, use short excerpts. Ask the user if they have the creator's permission when the clip is long. ## How a pro cuts this | beat | % | what happens | | --- | --- | --- | | hook | 0-10 | the most surprising moment of the source + your strongest reaction, in the first 1.5 s | | context | 10-20 | who / what this is, one line, credit on screen | | source + reactions | 20-80 | play the source; pause (freeze) or zoom where you talk over it | | verdict | 80-95 | your take: the value you add | | CTA | last 2 s | follow / part 2 | - **Sync is sacred:** do NOT run autocut on a reaction (`--no-cut`); every removed pause would desync your face from the source. Cut only whole sections with `--cut` ranges and move the source the same amount (a new clip element with its own `clipOffset`). - **Pacing:** ASL 2-4 s; a change every 3 s: zoom on the source detail, freeze-frame + callout, a punch-in on your face. - **Audio:** source audio plays as a bed ducked ~10 dB whenever you speak (the plan `music` block does this); your voice leads. - **Transitions:** cuts only; flash on the reveal moment at most once. - **Colour:** correct the facecam only (`color` affects the speaker video); leave the source as is. - **Captions:** your words only (the source can carry its own), preset `boxed` or `hormozi`, in the band between the two panes. - **Client judges:** is the reaction genuine, does the commentary add value, hold rate. - **Common mistakes:** drifting sync, source louder than the reactor, no credit, reacting with nothing to say, captions over the source's own subtitles. ## Make the edit ```bash python CORE/scripts/run.py facecam.mp4 --style SKILLS/split-proof --genre reaction-duet --no-cut \ --clip source.mp4 --format 9:16 --work react --stop-after plan python CORE/scripts/render.py react/plan.json --frames 1,4,8,14 ``` `--clip source.mp4` makes the source available to elements as clip id `main`. Then wire the layout and the audio in the plan (next section), review and render: ```bash python CORE/scripts/editor.py review react/plan.json --genre reaction-duet --fix python CORE/scripts/render.py react/plan.json --frames 1,4,8,14 # LOOK: source visible, face visible, credit python CORE/scripts/render.py react/plan.json python CORE/scripts/finish.py react/plan.json --platform tiktok ``` ## Refine the plan - **Source pane (split, top half):** `{"type": "clip", "clip": "main", "clipOffset": OFFSET, "start": 0, "end": DUR, "x": 540, "y": 480, "w": 1080, "h": 960, "in": "none"}`. The facecam already fills the bottom half in `split-proof`. - **PiP instead:** use `halo-cut` (face full frame) and a smaller source: `"w": 620, "h": 350, "x": 540, "y": 520, "className": "glass"`; or swap: source full (`w`/`h` = frame) and set `"speaker": {"mode": "panel", "rect": [700, 1300, 330, 440], "radius": 40}` for the face. - **Source audio, ducked:** `"music": {"src": "../source.mp4", "start": OFFSET, "volume": 0.5, "duck": true, "duckDepth": 10, "fadeOut": 0.5}`. `start` is the source second at video 0, so it matches `clipOffset`. - **Freeze on a moment:** end the playing clip element at T and add an `image` element (a still exported with `ffmpeg -ss <src time> -i source.mp4 -frames:v 1 freeze.png`) from T to T+2, then a new clip element with `clipOffset` = OFFSET + T + 2 (the source jumps past what was frozen; adjust `music.start` the same way only if you split the audio too). - **Callouts:** `{"type": "callout", "point": [700, 420], "x": 300, "y": 180, "title": "watch his hand"}`, `scribble` circles on the detail. - **Credit:** `{"type": "chip", "text": "Source: @creator", "start": 0, "end": DUR, "x": 60, "y": 120, "anchor": "top-left"}`. - **Captions:** `{"type": "captions", "preset": "boxed", "y": 1000, "dodgeFace": false}` - place between panes. ## Deliver ```bash python CORE/scripts/deliver.py react/plan.json --to tiktok,reels,shorts --out-dir react/deliver python CORE/scripts/qc.py react/deliver/final_tiktok.mp4 --plan react/plan.json --platform tiktok ``` - Put the original link in `post.md`'s description. Platform-native duet/stitch features keep attribution automatically; this edit is for re-uploads with commentary. ## Check before delivering - Lips and the source stay in sync at the start, middle and end (compare 3 stills). - Reactor always intelligible; source audible in gaps; no clipping (`qc.py` -14 LUFS / -1 dBTP). - Credit on screen the whole time; commentary present in every section. - Captions don't cover the source's own text or the face. - Tell the user: layout, offset used, length, file paths, and the rights caveat. ## Example requests - "Make a reaction video: this clip on top, my face below, duck the clip when I talk." - "Stitch-style: first 3 seconds of their video, then my answer." - "PiP my facecam over this trailer, freeze when I explain the shot."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.