ad-ladder
Ad ladder editor for Claude: turns one commercial shoot (talent takes, product shots, supers, a licensed track) into a full ad package - a 60 s or 30 s story master, a 15 s cutdown and a rebuilt 6 s bumper, each frame-exact, in 16:9, 1:1, 4:5 and 9:16, captioned and textless, mix
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/ad-ladder
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
Ad Ladder
One idea, told at every length: the 30 sells the story, the 15 sells the benefit, the 6 sells the brand.
- Best for: TV and online commercials, paid social campaigns, YouTube pre-roll and bumpers, product launches
- Inputs: script/board, talent and product takes, logo PNG, legal supers text, brand colours/fonts, licensed music
- Default skin:
kinetic-launch· alternativesproduct-spotlight,minimal-mono,ugc-ad-kit(selfie-style ads) - Formats: 16:9 master; 1:1, 4:5 and 9:16 versions of every length
- Durations (frame-exact): 60 s, 30 s, 15 s, 6 s bumper (YouTube bumpers are non-skippable and max 6 s)
- Genre preset:
--genre ad-ladder(python CORE/scripts/editor.py genre ad-ladder)
Before you start
- Needs video-studio-core in the same skills folder.
CORE= that folder,SKILLS= the skills folder. - First run on a machine:
python CORE/scripts/check_env.pyand fix what it reports. - Ask for: the single-minded proposition, the mandatory supers (legal lines, offer terms, URL), where each version runs (TV, YouTube, Meta, TikTok), the music licence terms (lengths and territories), and the brand guide.
- Claims, prices and offers come only from the client's approved script. Never invent a price, a rating or a result.
How a pro cuts this
The ladder is rebuilt, not trimmed. Each length has its own structure. Cutting the 30 down to 6 by trimming gives a bumper with no idea in it; build the 6 around one image, one line and the logo.
| length | structure | rule of thumb |
|---|---|---|
| 60 s | hook 0-3 s → situation → tension → product as the turn → proof → offer → logo + CTA | room for a mini-story and a second proof beat |
| 30 s | hook 0-2 s → problem → product → benefit shown → offer/CTA → logo (last 3-4 s) | one story, one proof |
| 15 s | hook 0-1.5 s → benefit shown → product → logo + CTA (last 3 s) | drop the setup, keep the payoff |
| 6 s | one striking image → product/benefit line → logo | one idea; brand visible by 2 s, logo on screen ≥ 1.5 s |
Pacing. ASL 1-2.5 s; a visual change every ~2 s; the 6 s holds 2-4 shots. Product and logo shots get the longest holds. Brand visible early (within the first 5 s of a 30, 2 s of a 6) for skippable placements.
Cut rules. Cut on action and motion; match cuts between product moments; smash cut into the reveal. Picture leads, VO follows: J-cut voice under the incoming shot by 2-4 frames. No dead frames at the head or tail.
Frame-exact lengths. A 30 is 30.00 s (900 frames at 30 fps), not 29.7 or 30.4. Check each output with qc.py
(duration row) and fix by adjusting the last shot, never by speeding the whole spot.
Supers (legal text). Readable: hold at least 3 s plus 1 s per extra line (about 15 cps), inside title-safe, contrast against the footage. A super that flashes by fails legal review.
Music and sound. Track leads (bed about -18 dB, ducked about 8 dB under VO); every length ends on a resolved bar
(music_fit.py per length, "endButton": true). Logo sting or a soft hit on the logo; whooshes only on real motion.
Colour. Product true to its real colour first (match to a reference frame of the product), then a light look:
clean-commercial, food-pop (food), cool-corporate (tech), intensity 0.4-0.6. Skin protected.
Captions. Social versions get burned-in captions ("preset": "clean" or the skin's style), ≤ 15 cps;
TV versions get none on picture and a separate caption file.
What the client judges. Product looks right, logo time, legal supers readable, every version delivered at the exact length and spec, the message survives in the 6 s.
Common mistakes. Bumper trimmed from the 30; logo appearing in the last 0.5 s only; supers too short to read; spots a few frames long or short; music cut mid-phrase; vertical versions with the product cropped off.
Make the edit
# 1. selects from the shoot (takes, retakes, best performance)
python CORE/scripts/selects.py --work sel --input shoot.mp4
# tick paper_edit.md, then build the 30 s master first
python CORE/scripts/selects.py --work sel --apply --format 16:9
python CORE/scripts/run.py - --timeline sel/selects_spec.json --style SKILLS/kinetic-launch --genre ad-ladder --format 16:9 --no-cut --music track.wav --logo logo.png --duration 30 --work ad30 --stop-after plan
python CORE/scripts/editor.py review ad30/plan.json --genre ad-ladder --fix
python CORE/scripts/render.py ad30/plan.json --frames 0.5,5,12,20,27,29.5
python CORE/scripts/render.py ad30/plan.json
python CORE/scripts/finish.py ad30/plan.json --platform youtube
Each other length gets its own paper edit and work folder. Re-order the shoot for the length with
story.py outline --work sel / story.py reorder --work sel --order "..." --format 16:9 --out ad15.json (or write a
timeline.py spec by hand: see CORE/references/timeline.md), then the same run with --duration 15 / --duration 6.
Cut music to each length on the beat grid:
python CORE/scripts/music_fit.py track.wav --duration 15 --out track_15.wav
python CORE/scripts/music_fit.py track.wav --duration 6 --out track_06.wav
Aspect versions: build a vertical master per length with --format 9:16 (re-frame each shot with
"reframe": "face" in the timeline spec, or check product framing by hand), or let deliver.py reframe the 16:9.
Always check the product and logo are still in frame in the 9:16 and 4:5 contact sheets.
Textless: copy plan.json to plan_textless.json, delete every text/captions/card element (keep the picture and
mix), render to a separate composite and finish with --out ad30_textless.mp4.
Refine the plan
- Hook: first shot is the strongest image, no logo slate. Set the
hooktext element (≤ 6 words) or remove it if the picture carries the hook. - Supers:
{"type": "text", "text": "Offer valid till 31 Dec. T&Cs apply.", "start": 25.5, "end": 29.9, "y": "=H*0.88", "w": "=W*0.8", "style": {"fontSize": "30px"}}- hold ≥ 3 s; keep inside title-safe; never over the product.
- Logo + CTA:
{"type": "image", "image": "{logo}", "start": 26, "end": 30, "in": "springPop", "w": 520}and achipwith the URL; land the logo on a downbeat ("music": {"fitTo": "beats", "endButton": true}). - Product callouts:
calloutorpointerelements fromreferences/motion.mdon the feature being shown. - Price/offer:
counteronly for approved numbers; checkprefix/suffix. - Colour:
"color": {"look": "clean-commercial", "intensity": 0.5, "match": "product_ref.png"}. - Audio:
"audio": {"platform": "youtube", "voice": {"denoise": "light", "eq": "broadcast"}},"music": {"src": "track_30.wav", "under": 18, "duck": "speech-aware", "endButton": true}. - Length: set
"output": {"duration": 30}exactly; trim or extend the last shot in the timeline spec to match.
Deliver
python CORE/scripts/deliver.py ad30/plan.json --to youtube,reels,tiktok,facebook,prores-422hq --out-dir ad30/deliver
python CORE/scripts/deliver.py ad06/plan.json --to youtube,shorts --out-dir ad06/deliver
python CORE/scripts/qc.py ad30/deliver/final_youtube.mp4 --plan ad30/plan.json --platform youtube --strict
TV: finish with --platform broadcast-us (-24 LKFS, ATSC A/85) or --platform broadcast-eu (-23 LUFS, EBU R128) and
deliver broadcast-xdcam or prores-422hq (deliver.py --list shows all presets). Name every file with
length, aspect and version: brand_spring_30s_16x9_v002.mp4, ..._textless.mp4.
Check before delivering
- Every length is frame-exact (
qc.pyduration row) and every aspect exists for every length that needs it. - The 6 s has one idea, the brand by 2 s, the logo ≥ 1.5 s.
- Supers readable (≥ 3 s, ≤ 15 cps), inside title-safe, legally approved text only.
- Product colour true; product and logo in frame in vertical versions.
- Loudness per destination (-14 LUFS social, -24 LKFS US TV, -23 LUFS EU TV), true peak within limits; music ends on a bar.
- Textless versions and the caption files delivered alongside.
Example requests
- "Here's our commercial shoot and script: cut a 30, a 15 and a 6-second bumper, with vertical versions for Reels and TikTok."
- "Make the TV version of this 30-second spot at broadcast loudness, plus a textless master."
- "Rebuild a 6-second YouTube bumper from this ad: one idea, logo on screen at the end."
Files (claude-video-editing-skills)
-
SKILL.md 8.9 KB
--- name: ad-ladder description: "Ad ladder editor for Claude: turns one commercial shoot (talent takes, product shots, supers, a licensed track) into a full ad package - a 60 s or 30 s story master, a 15 s cutdown and a rebuilt 6 s bumper, each frame-exact, in 16:9, 1:1, 4:5 and 9:16, captioned and textless, mixed to platform or broadcast loudness. Use when the user says 'cut a 30 and a 15', 'make ad cutdowns', 'we need a 6 second bumper', 'versions for every placement', 'TV spot and social versions', 'ad package', or hands over a commercial shoot with a script. Runs on video-studio-core (selects, story, timeline, colour, mix, deliver, QC)." license: MIT --- # Ad Ladder One idea, told at every length: the 30 sells the story, the 15 sells the benefit, the 6 sells the brand. - **Best for:** TV and online commercials, paid social campaigns, YouTube pre-roll and bumpers, product launches - **Inputs:** script/board, talent and product takes, logo PNG, legal supers text, brand colours/fonts, licensed music - **Default skin:** `kinetic-launch` · alternatives `product-spotlight`, `minimal-mono`, `ugc-ad-kit` (selfie-style ads) - **Formats:** 16:9 master; 1:1, 4:5 and 9:16 versions of every length - **Durations (frame-exact):** 60 s, 30 s, 15 s, 6 s bumper (YouTube bumpers are non-skippable and max 6 s) - **Genre preset:** `--genre ad-ladder` (`python CORE/scripts/editor.py genre ad-ladder`) ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = that folder, `SKILLS` = the skills folder. - First run on a machine: `python CORE/scripts/check_env.py` and fix what it reports. - Ask for: the single-minded proposition, the mandatory supers (legal lines, offer terms, URL), where each version runs (TV, YouTube, Meta, TikTok), the music licence terms (lengths and territories), and the brand guide. - Claims, prices and offers come only from the client's approved script. Never invent a price, a rating or a result. ## How a pro cuts this **The ladder is rebuilt, not trimmed.** Each length has its own structure. Cutting the 30 down to 6 by trimming gives a bumper with no idea in it; build the 6 around one image, one line and the logo. | length | structure | rule of thumb | | --- | --- | --- | | 60 s | hook 0-3 s → situation → tension → product as the turn → proof → offer → logo + CTA | room for a mini-story and a second proof beat | | 30 s | hook 0-2 s → problem → product → benefit shown → offer/CTA → logo (last 3-4 s) | one story, one proof | | 15 s | hook 0-1.5 s → benefit shown → product → logo + CTA (last 3 s) | drop the setup, keep the payoff | | 6 s | one striking image → product/benefit line → logo | one idea; brand visible by 2 s, logo on screen ≥ 1.5 s | **Pacing.** ASL 1-2.5 s; a visual change every ~2 s; the 6 s holds 2-4 shots. Product and logo shots get the longest holds. Brand visible early (within the first 5 s of a 30, 2 s of a 6) for skippable placements. **Cut rules.** Cut on action and motion; match cuts between product moments; smash cut into the reveal. Picture leads, VO follows: J-cut voice under the incoming shot by 2-4 frames. No dead frames at the head or tail. **Frame-exact lengths.** A 30 is 30.00 s (900 frames at 30 fps), not 29.7 or 30.4. Check each output with `qc.py` (duration row) and fix by adjusting the last shot, never by speeding the whole spot. **Supers (legal text).** Readable: hold at least 3 s plus 1 s per extra line (about 15 cps), inside title-safe, contrast against the footage. A super that flashes by fails legal review. **Music and sound.** Track leads (bed about -18 dB, ducked about 8 dB under VO); every length ends on a resolved bar (`music_fit.py` per length, `"endButton": true`). Logo sting or a soft hit on the logo; whooshes only on real motion. **Colour.** Product true to its real colour first (match to a reference frame of the product), then a light look: `clean-commercial`, `food-pop` (food), `cool-corporate` (tech), intensity 0.4-0.6. Skin protected. **Captions.** Social versions get burned-in captions (`"preset": "clean"` or the skin's style), ≤ 15 cps; TV versions get none on picture and a separate caption file. **What the client judges.** Product looks right, logo time, legal supers readable, every version delivered at the exact length and spec, the message survives in the 6 s. **Common mistakes.** Bumper trimmed from the 30; logo appearing in the last 0.5 s only; supers too short to read; spots a few frames long or short; music cut mid-phrase; vertical versions with the product cropped off. ## Make the edit ```bash # 1. selects from the shoot (takes, retakes, best performance) python CORE/scripts/selects.py --work sel --input shoot.mp4 # tick paper_edit.md, then build the 30 s master first python CORE/scripts/selects.py --work sel --apply --format 16:9 python CORE/scripts/run.py - --timeline sel/selects_spec.json --style SKILLS/kinetic-launch --genre ad-ladder --format 16:9 --no-cut --music track.wav --logo logo.png --duration 30 --work ad30 --stop-after plan python CORE/scripts/editor.py review ad30/plan.json --genre ad-ladder --fix python CORE/scripts/render.py ad30/plan.json --frames 0.5,5,12,20,27,29.5 python CORE/scripts/render.py ad30/plan.json python CORE/scripts/finish.py ad30/plan.json --platform youtube ``` **Each other length** gets its own paper edit and work folder. Re-order the shoot for the length with `story.py outline --work sel` / `story.py reorder --work sel --order "..." --format 16:9 --out ad15.json` (or write a `timeline.py` spec by hand: see `CORE/references/timeline.md`), then the same run with `--duration 15` / `--duration 6`. Cut music to each length on the beat grid: ```bash python CORE/scripts/music_fit.py track.wav --duration 15 --out track_15.wav python CORE/scripts/music_fit.py track.wav --duration 6 --out track_06.wav ``` **Aspect versions:** build a vertical master per length with `--format 9:16` (re-frame each shot with `"reframe": "face"` in the timeline spec, or check product framing by hand), or let `deliver.py` reframe the 16:9. Always check the product and logo are still in frame in the 9:16 and 4:5 contact sheets. **Textless:** copy `plan.json` to `plan_textless.json`, delete every text/captions/card element (keep the picture and mix), render to a separate composite and finish with `--out ad30_textless.mp4`. ## Refine the plan - **Hook:** first shot is the strongest image, no logo slate. Set the `hook` text element (≤ 6 words) or remove it if the picture carries the hook. - **Supers:** `{"type": "text", "text": "Offer valid till 31 Dec. T&Cs apply.", "start": 25.5, "end": 29.9, "y": "=H*0.88", "w": "=W*0.8", "style": {"fontSize": "30px"}}` - hold ≥ 3 s; keep inside title-safe; never over the product. - **Logo + CTA:** `{"type": "image", "image": "{logo}", "start": 26, "end": 30, "in": "springPop", "w": 520}` and a `chip` with the URL; land the logo on a downbeat (`"music": {"fitTo": "beats", "endButton": true}`). - **Product callouts:** `callout` or `pointer` elements from `references/motion.md` on the feature being shown. - **Price/offer:** `counter` only for approved numbers; check `prefix`/`suffix`. - **Colour:** `"color": {"look": "clean-commercial", "intensity": 0.5, "match": "product_ref.png"}`. - **Audio:** `"audio": {"platform": "youtube", "voice": {"denoise": "light", "eq": "broadcast"}}`, `"music": {"src": "track_30.wav", "under": 18, "duck": "speech-aware", "endButton": true}`. - **Length:** set `"output": {"duration": 30}` exactly; trim or extend the last shot in the timeline spec to match. ## Deliver ```bash python CORE/scripts/deliver.py ad30/plan.json --to youtube,reels,tiktok,facebook,prores-422hq --out-dir ad30/deliver python CORE/scripts/deliver.py ad06/plan.json --to youtube,shorts --out-dir ad06/deliver python CORE/scripts/qc.py ad30/deliver/final_youtube.mp4 --plan ad30/plan.json --platform youtube --strict ``` TV: finish with `--platform broadcast-us` (-24 LKFS, ATSC A/85) or `--platform broadcast-eu` (-23 LUFS, EBU R128) and deliver `broadcast-xdcam` or `prores-422hq` (`deliver.py --list` shows all presets). Name every file with length, aspect and version: `brand_spring_30s_16x9_v002.mp4`, `..._textless.mp4`. ## Check before delivering - Every length is frame-exact (`qc.py` duration row) and every aspect exists for every length that needs it. - The 6 s has one idea, the brand by 2 s, the logo ≥ 1.5 s. - Supers readable (≥ 3 s, ≤ 15 cps), inside title-safe, legally approved text only. - Product colour true; product and logo in frame in vertical versions. - Loudness per destination (-14 LUFS social, -24 LKFS US TV, -23 LUFS EU TV), true peak within limits; music ends on a bar. - Textless versions and the caption files delivered alongside. ## Example requests - "Here's our commercial shoot and script: cut a 30, a 15 and a 6-second bumper, with vertical versions for Reels and TikTok." - "Make the TV version of this 30-second spot at broadcast loudness, plus a textless master." - "Rebuild a 6-second YouTube bumper from this ad: one idea, logo on screen at the end."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.