Source: raw/Claude_Code_Just_Changed_Social_Media_Carousels_Forever.md Creator: Simon Scrapes (YouTube) · URL: https://www.youtube.com/watch?v=REEzMNF54GY · Duration: 17:44 · Uploaded: 2026-05-27 · Platform: YouTube

A walkthrough of Simon Scrapes’ “social content creator” skill system, which turns one input (a topic, video, URL, PDF or audio file) into a researched LinkedIn or Instagram carousel that follows a brand design system while varying slide layouts. He explains every stage so you can rebuild it, and sells the finished system in his paid community. The two load-bearing pieces are the visual identity files and a human review step before any images are generated.

Key Takeaways

  • Why carousels. He relays a report on 1.5 million LinkedIn posts finding carousels get about three times the reach of text posts, attributed to dwell time (15 to 20 seconds against 8 to 10), and says carousels had the highest Instagram engagement rate going into 2026, with a second distribution window. He gives 8 to 10 slides as the usual sweet spot. None of these are linked.
  • What makes AI carousels look like AI. No continuity between posts, identical slides, no hierarchy, no research, and irrelevant images or emoji.
  • Consistency is enforced, variety is deliberate. Fonts, accent colours and grids come from one tokens file every time; layouts vary between slides (separate hero, body and CTA templates) so readers don’t get bored.
  • First-run onboarding takes 5 to 25 minutes. It checks for a voice profile (or runs the brand voice skill), a visual identity (or builds one: palette, typography, templates), and writes a config file with default platform, language and format (carousel, single image or text).
  • Templates come from references. Give it three to five images you admire (Instagram posts, Canva or Figma templates, brand books). It rebuilds them as templates in your colours. He also had it generate full brand guidelines (logo usage, palette combinations, type scale, mock-ups), reverse-engineered from a branding agency’s deliverable he paid for.
  • Inputs it accepts. Your own video, a topic (triggers a trending-research skill across Reddit, X and the web), a feed scrape for ideas, a web article URL, a local audio file (transcribed), a PDF, or an existing post to adapt to another platform.
  • The pipeline. Check config → classify the input scenario → gather research → ask the objective (leads or teaching), format and platform → a designer subagent builds a visual inventory (logos, icons, photos, screenshots it can use), chooses a narrative arc and stress-tests the slide-one hook → outline each slide visually, then write headlines → human review of the caption and slide plan → an image-generator subagent with its own context produces each slide → an HTML preview mocking the LinkedIn or Instagram post.
  • Three ways a slide gets made: pure HTML from a template (no image model), an AI image inside an HTML template (it pulled his headshot and the Anthropic logo automatically), or real screenshots it captures from a URL you give it. A full carousel takes about one to two minutes.
  • The last 5% by hand. Import a generated slide into Canva Magic Layers to turn it into editable layers and fix text size or logo placement.
  • When not to bother. For an occasional one-off slide he says just use Canva; the system pays off when you post carousels weekly across platforms.

Implementation

Tool/Service: Claude Code skill system with designer and image-generator subagents; OpenAI image generation (captioned “GPT-2.0”, likely GPT Image 2) or Gemini; Apify for LinkedIn scraping; a social scheduling tool with an API (captioned “Zerno”) for drafts. Setup: Brand voice and visual identity files first (Brand Context Files), then a config file per brand. Multiple brands or clients are supported. Cost: Image-model and scraper API keys are your own; no totals given. Integration notes: The review gate after the slide plan sits before the step that costs money (image generation). He later describes the same system posting only after approval in a custom UI (Advice Reset).

Try It

  1. Build or collect your visual identity tokens and three to five reference carousels you like.
  2. Ask Claude Code to turn the references into hero, body and CTA HTML templates using your tokens.
  3. Write a skill that takes a topic, researches it, and returns a caption plus a slide plan (role, headline, body, image idea per slide) and stops for review.
  4. Only after approval, have a separate subagent render each slide from the plan.
  5. Fix the final details in Canva Magic Layers rather than re-prompting.

Open Questions

  • The carousel performance statistics are relayed without sources.
  • The scheduling tool name (“Zerno”) and the exact image model are caption-level; not verified.
  • No cost per carousel is given.