carousel-creator
Skill18 starsChanged 4 months agoMITView on GitHub
Create high-impact Instagram carousel posts with branded slides. Use when: (1) user says /carousel or 'create a carousel', (2) user provides a topic, YouTube transcript, or lead magnet and wants it turned into a carousel, (3) user wants branded carousel slides for Instagram, (4) user has existing video clips and wants a video carousel with branded text overlays, (5) user says 'top repos this week/month' or 'trending GitHub repos' or provides a github.com repo URL. Handles: content writing, GitHub trending/repo analysis, image generation, text overlay rendering, video overlay burning via FFmpeg, and final slide output.
- Installs packages
In this repository52 other files it gives its agents.
- ad-studioSkill
- brainstormingSkill
- build-siteSkill
- cold-outreachSkill
- content-engineSkill
- content-promptSkill
- daily-digestSkill
- daily-post-ideasSkill
- defuddleSkill
- digital-guide-builderSkill
- dissectSkill
- dm-writerSkill
- edu-siteSkill
- faceless-videoSkill
- first-customer-closerSkill
- frontend-designSkill
- higgsfield-generateSkill
- hooksSkill
- ig-channel-remixSkill
- infographicSkill
- launch-planSkill
- lead-magnetSkill
- linkedin-post-batchSkill
- linkedin-postSkill
- meeting-notesSkill
- money-model-auditSkill
- objection-handlerSkill
- offer-builderSkill
- one-lead-magnetSkill
- outlinesSkill
- personalizeSkill
- positioning-filterSkill
- pricing-calculatorSkill
- productize-serviceSkill
- proposal-generatorSkill
- saas-ad-studioSkill
- sanitize-projectSkill
- shortifySkill
- substackSkill
- ui-ux-pro-maxSkill
- viral-format-minerSkill
- viral-remixSkill
- warm-outreachSkill
- web-design-guidelinesSkill
- yapSkill
- youtube-clipping-agentSkill
- yt-descriptionSkill
- yt-introSkill
- yt-searchSkill
- yt-thumbnailSkill
- yt-titlesSkill
- skill-nameSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 9 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Create branded Instagram carousel posts from topics, YouTube transcripts, lead magnets, or GitHub repos.
- Distinctive: line 16
If it doesn't exist, run the personalize skill first (or just tell me your name, niche, offer, and audience — I'll create it).
- Distinctive: line 17
If a field this skill needs is blank, I'll ask one or two quick questions and save the answers back to the profile so you only answer once.
- Distinctive: line 18
Never invent facts or results about the user — use only what's in the profile, or ask.
- Very specific: line 28
Never embed an API key in the repo.
- Distinctive: line 57
No API key needed. If captions are disabled, ask the user to paste the transcript.
- Very specific: line 82
If --count N is set, generate N independent copy sets (different hook + framework each), present all for approval, then render.
- Very specific: line 100
Read references/carousel-frameworks.md for slide-by-slide blueprints.
- Distinctive: line 123
Write 5-10 slides: minimum 5 (cover + 3 body + CTA), sweet spot 7-8, max 10.
- Distinctive: line 147
Self-check before rendering: read the slides as one paragraph.
- Distinctive: line 159
Use emoji (👉) instead of unicode arrows
- Distinctive: line 164
Follow the brand voice from your profile.
- Very specific: line 228
No JSON — slides are rendered individually by Higgsfield nano_banana_2 from per-slide prompts.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Carousel Creator
Create branded Instagram carousel posts from topics, YouTube transcripts, lead magnets, or GitHub repos.
Personalization
This skill works in your brand voice. Before running, load the brand profile at ~/.claude/brand-profile.md.
- If it doesn't exist, run the personalize skill first (or just tell me your name, niche, offer, and audience — I'll create it).
- If a field this skill needs is blank, I'll ask one or two quick questions and save the answers back to the profile so you only answer once.
- Never invent facts or results about the user — use only what's in the profile, or ask.
The profile supplies: your tone/voice, your offer (what you sell or your community), your default CTA, your booking/link destination, and your proof (a real win or number you can stand behind). Wherever this skill says "your offer", "your CTA", or "your proof", pull it from the profile — never hardcode.
Requirements
- Higgsfield CLI or MCP — for photo covers (
gen_cover.py) and the RISO/editorial render path. Graceful fallback: the default alternating-solid renderer (render_solid.py) needs no API and works fully offline. - ffmpeg — video-carousel mode only (
brew install ffmpeg). - Playwright MCP — optional, for GitHub repo screenshots.
- Pillow —
pip install Pillow(all rendering).
Never embed an API key in the repo. Higgsfield reads its own credentials from your shell/~/ environment.
Workflow
- Parse input — identify source type and extract content
- Select style — solid (default), photo overlay, RISO, or video
- Select framework — choose a carousel blueprint
- Write slides — slide-by-slide copy in your voice
- Generate the carousel JSON — structured spec for the renderer
- Render slides — run the appropriate renderer
- Publish / hand off — slides saved locally; you post or schedule them
Step 1: Parse Input
Determine input type:
- Topic/idea — short phrase → research with WebSearch first
- YouTube URL — fetch transcript (below), then analyze
- Pasted transcript — extract key insights, story, framework
- Lead magnet — the carousel promotes an existing free asset
- GitHub trending / repo URL — see
references/github-mode.md
YouTube Transcript Extraction
pip install -q youtube-transcript-api 2>/dev/null
python3 -c "
from youtube_transcript_api import YouTubeTranscriptApi
t = YouTubeTranscriptApi.get_transcript('VIDEO_ID')
print(' '.join(x['text'] for x in t))
"No API key needed. If captions are disabled, ask the user to paste the transcript.
Content Analysis (for transcripts)
Extract before writing slides:
- Main thesis — one big idea, one sentence
- 2-3 distinct angles — each a standalone teaching moment
- 3-6 quotable moments per angle — short, punchy phrases
- Audience pain point — what problem this solves
- Emotional hook per angle — the tension or feeling
Present angles to the user for approval before writing slides.
CTA mode
- GitHub / lead-magnet carousel → CTA = "Comment [KEYWORD] and I'll DM you [the thing]". Pick a random, memorable, unrelated keyword (e.g.
WAFFLE,TORCH,SUMMIT) — never a topic word. If you use a comment-to-DM tool, check that keyword isn't already live in it. - Everything else → CTA = your default CTA / offer (from the brand profile). "Link in bio" on the slide; the URL goes in the caption only.
Display before writing:
Carousel: {summary}
Style: {solid / photo / riso / video}
Framework: {chosen framework}
CTA: {your offer OR lead_magnet — KEYWORD}
Slides: {count}
Est. cost: $0.00 (solid) or ~${count * 0.13} (photo)If --count N is set, generate N independent copy sets (different hook + framework each), present all for approval, then render.
Step 2: Select Style
| Style | Renderer | Cost | Best for |
| Alternating solid (default) | scripts/render_solid.py | $0.00 | How-tos, step-by-steps, informational, quick turnaround |
| Photo overlay | scripts/generate_carousel.py | ~$0.13/slide | Storytelling, personal brand, emotional impact |
| RISO / editorial poster | Higgsfield nano_banana_2 | ~$0.03/slide | Brand-forward print aesthetic — see references/riso-mode.md |
| Video clips | scripts/generate_video_carousel.py | $0.00 | You have existing footage — burns branded overlay per clip |
Default to alternating solid for educational content. Use photo overlay for storytelling, RISO when asked for "poster/editorial/riso" style, video when the user supplies clips.
Alternating solid uses bold type on dark backgrounds (#0A0A0A / #1A1A1A / #141414) with a green accent, plus a warm-cream light_pop mode (#E8DCC4 bg, charcoal text, orange accent). No image generation. See references/brand-style.md.
GitHub source carousels: solid body slides + a photo cover via gen_cover.py. See references/github-mode.md.
Step 3: Select Framework
Read references/carousel-frameworks.md for slide-by-slide blueprints. Choose ONE:
| Framework | Best for |
| Checklist / Audit | Actionable lists, "X things you need" |
| Myth vs Fact | Challenging beliefs, contrarian takes |
| Before / After (Reframe) | Transformations, "generic vs specific" |
| Step-by-step | How-to guides, processes, systems |
| Teardown | Analyzing what works, case studies |
Step 4: Write Slides
Hook first (always)
Lock the slide-1 hook before anything else. Start from your saved hooks if you have a swipe file, or write a fresh one.
Formula: mechanism + outcome in one line. The mechanism is the tool or system; the outcome is the result or tension. Both must be on slide 1.
- Bad: "Your AI marketing team starts here" — no mechanism, no outcome
- Bad: "One command" — mechanism only
- Good: "I went from 3 hours per post to 60 seconds with one command" — transformation + mechanism
The hook names the result and the tool that delivers it. It is not a teaser or a mystery. If the hook is weak, the carousel is dead — fix it before writing body slides.
Length
Write 5-10 slides: minimum 5 (cover + 3 body + CTA), sweet spot 7-8, max 10. Every slide earns its place or gets cut. Never pad.
Narrative first — one big idea per slide (non-negotiable)
A carousel is a STORY, not a deck. Write the narrative spine first: the single sentence each slide adds. Each slide carries ONE big idea that only makes sense after the slide before it. If you can shuffle the slides without breaking the flow, rewrite.
Every reader should feel themselves LEVELING UP — each slide gives a sharper diagnosis, a hidden cause, a move, a system. By the final value slide they should feel a distinct unlock: "oh, THAT'S what was stopping me."
Two failure modes to avoid:
- Restating — two slides saying the same thing. Cut one.
- List soup — every slide a bullet list. Only ONE slide may be a list: the System slide. Every other slide is one concrete idea in prose.
Be clear and concrete. Tie it to money and stakes, not vague abstractions.
Default narrative spine (adapt, keep the escalation):
- Hook — one sentence, under 12 words. Mechanism + outcome/tension.
- The real problem — concrete, money-tied. The misdiagnosis.
- Why it fails — the mechanism of the pain; make them feel the cost.
- The turn — what you did instead. Lead with a hard result/number when you have one.
- The System — the ONE allowed list. Numbered steps of the real build.
- The payoff / seed — relate the system to THE READER's own situation; make the gap visible without selling.
- CTA — hard CTA based on mode.
For shorter carousels merge adjacent beats, but never merge problem and payoff, and never duplicate a beat.
Self-check before rendering: read the slides as one paragraph. Does each line need the one before it? Anything repeat? More than one list? If yes, rewrite.
Writing rules
- 6th-grade reading level — clarity, not complexity
- Specific > generic — name the real outcome, tool, and number
- Speak from experience, not at people; short lines, one thought per line
- Double line break (
\n\n) between every sentence in body copy - Vary vocabulary across slides — don't repeat the same core noun on every slide
- Every claim must make literal sense — read each line cold, fix logic
- Proof slides: use an in-copy connector for action→result (e.g. "I GAVE CLAUDE MY LIST = 1,154 REAL LEADS"). Keep the swipe arrow isolated bottom-right.
- Mark ONE key phrase per slide for accent:
{green}phrase{/green}(photo) or theaccent_phrasefield (solid) - Wrap italic lines in
*...* - Use emoji (👉) instead of unicode arrows
- Numbered lists are for the System slide ONLY
- Max 4-5 body lines per slide — split dense content into another slide
Copy rules
- Follow the brand voice from your profile.
- Use only real proof — pull your proof/number from the profile; never invent a follower count, revenue figure, or member count.
- CTA slide → "Link in bio" (offer mode) or "Comment [KEYWORD]" (lead-magnet mode). Never put a raw URL on the slide; URL goes in the caption.
- Pick ONE CTA line / tagline from your profile that fits the carousel's topic.
Step 5: Generate Carousel JSON
Solid / photo spec
Save to output/carousel_spec.json:
{
"title": "Internal reference title",
"style": "alternating_solid",
"cta_mode": "offer",
"cta_keyword": null,
"cta_offer": null,
"slides": [
{ "number": 1, "bg_mode": "dark", "headline": "HOOK IN CAPS", "body": "", "accent_phrase": "KEY WORD" },
{ "number": 2, "bg_mode": "light", "headline": "SLIDE HEADLINE", "body": "Line one.\n\nLine two.", "accent_phrase": "phrase" },
{ "number": 7, "bg_mode": "gradient", "show_logo": true, "headline": "WANT THE SYSTEM?", "body": "Link in bio.", "cta_button": "Get it" }
]
}- For photo overlay, use
shot_type+image_prompt(generate, ~$0.13) orlibrary_image(reuse, free) per slide, and{green}...{/green}accent markers. Seereferences/brand-style.md. - For
cta_mode: "lead_magnet", setcta_keywordandcta_offer. - Full field list for the solid renderer (bg modes,
stat_value,cta_button,show_bookmark,github_card,bottom_summary,image_path,dark_card_text,cover_image_path,topic_label,step_label): seereferences/brand-style.mdandreferences/github-mode.md.
Video carousel spec
Save to output/video_carousel_spec.json. Use clip_path (bare filename resolved from BROLL_DIR, or an absolute path):
{
"title": "…", "cta_mode": "offer",
"slides": [
{ "number": 1, "type": "cover", "clip_path": "clip1.mp4", "headline": "HOOK", "body": "One-line payoff", "green_accent": "key phrase" }
]
}bodyis required on every slide (a one-line overlay looks thin). Headline ALL CAPS, onegreen_accentper slide. Audio is stripped; pre-trim clips.
Step 6: Render Slides
Scripts live in scripts/. Output goes to output/. Set CAROUSEL_ROOT if you run from elsewhere (defaults to the skill root).
Alternating solid (default, $0.00, offline)
cd <skill>/scripts
python3 render_solid.py ../output/carousel_spec.jsonProgress bar + swipe chevron are automatic (last slide gets no chevron). Slides save to output/slides/slide_XX.jpg + manifest.json. To draw your logo lockup on show_logo slides, set BRAND_LOGO (image path) or BRAND_NAME (wordmark text).
Photo overlay (~$0.13/slide)
Show the cost estimate first (only slides needing new generation count). Then:
CAROUSEL_ROOT="<skill>" python3 scripts/generate_carousel.py output/carousel_spec.jsonGenerates images via the Higgsfield backend, applies branded overlays, saves raw images to a local library for reuse, and writes final slides to output/slides/. For a character (CTA face) shot, set BRAND_FACE or BRAND_PORTRAIT to a portrait image path — if unset, the shot generates with no reference.
Video clips ($0.00, needs ffmpeg)
BROLL_DIR="./b-roll/videos" python3 scripts/generate_video_carousel.py output/video_carousel_spec.jsonBurns a gradient + branded text overlay (0.9 brightness + vignette + drop shadow) onto each clip. Saves silent MP4s to output/video_slides/ + manifest.json.
RISO / editorial poster
No JSON — slides are rendered individually by Higgsfield nano_banana_2 from per-slide prompts. See references/riso-mode.md for the full workflow, prompt templates, and the parallel-render-order pattern.
Step 7: Publish / Hand Off
Rendering is local only — nothing is uploaded automatically.
- Show results — list
output/slides/(oroutput/video_slides/) in order, total cost, and ask: regenerate any slides? adjust copy? ready to post? - Write the caption — no hashtags unless requested; expand on the content, conversational, in your voice.
- Lead-magnet mode: open with "Comment [KEYWORD] and I'll send it to you."
- Offer mode: soft close at the very end — one line about your offer + "Link in bio." Use only real proof from your profile.
- Post or schedule — upload the slides in
output/slides/to your Instagram scheduler (Meta, Later, Buffer, or similar) or post manually. Slides are numberedslide_01…slide_NNso carousel order is preserved. - Lead magnet? — if you use a comment-to-DM keyword, set that keyword and its DM copy up in your comment-to-DM automation tool. The keyword lives in the caption, not on the slide.
Flags
| Flag | Default | What it does |
--count N | 1 | Generate N carousels on one topic, each a different hook + framework; all presented before rendering |
--slides N | auto (5-10) | Target a specific slide count |
--cta "text" | your default CTA | Custom CTA text for the final slide |
Reference Files
- references/brand-style.md — colors, typography, solid-renderer fields, image-prompt templates
- references/carousel-frameworks.md — slide-by-slide framework blueprints
- references/github-mode.md — trending-repos + single-repo teardown carousels
- references/riso-mode.md — editorial/RISO poster render workflow
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

