LinkedIn Carousel Maker

Drop in a LinkedIn post and a few reference images. The pipeline reads your design language from the references, rewrites your post as a 6-slide carousel (cover + 4 content slides + CTA), generates all slides at once at 928×1152px, then runs a full refinement pass via GPT Image 2. You get a complete, design-consistent carousel deck in minutes — built from your own content and visual style.

Real outputs

• 
  Cover
• 
  Slide 2
• 
  Slide 3
• 
  Slide 4
• 
  Slide 5
• 
  CTA

Live on the feed

linkedin.com/feed/

Alex Tanoa

Founder, InfuseLab · 2h

Turned one post into a 6-slide carousel in about a minute. Same argument, built for the scroll ↓

1 / 6

👍 ❤️ 💡 214 · 37 comments · 12 reposts

How it works

01

Your inputs

LinkedIn post

LinkedIn post (text or markdown) · Creative reference images (2–3) · Custom CTA text

02

Brand context

Your brand rules, palette, product facts, and approved references — applied to every prompt in the run.

03

Bounded generation

1. 01Reference Image Analysis
2. 026-Slide Content Plan
3. 03Prompt Extraction (Parallel)
4. 04Prompt Refinement
5. 05Initial Slide Generation (Parallel)
6. 06Quality Refinement Pass (Parallel)

GPT-4.1 · Claude Sonnet 4.6 · Claude Haiku 4.5 +2

04

Automated quality pass

Automated evaluation pass against the brief.

05

Human approval gate

Final output requires human approval before delivery.

06

Delivery

6 LinkedIn carousel slides (928×1152px portrait)

The bounded path every run follows — your inputs in, the real delivered output on the right. Step detail below.

1. 01

   Reference Image Analysis

   GPT-4.1 Vision reads your creative reference images and extracts the full design language: subjects, layout structure, visual treatment, colour palette, connector and arrow styles, and slide-ready design elements. This becomes the style anchor for every slide.

2. 02

   6-Slide Content Plan

   Claude Sonnet 4.6 reads your LinkedIn post alongside the visual analysis and generates a complete carousel spec as JSON — cover slide, 4 content slides, and a CTA slide. Each slide gets a headline, body copy, visual direction, design notes, and an image generation prompt.

3. 03

   Prompt Extraction (Parallel)

   Claude Haiku 4.5 extracts the image generation prompt for each of the 6 slides simultaneously — isolating clean, model-ready prompts from the broader JSON spec.

4. 04

   Prompt Refinement

   Claude Sonnet 4.6 runs a refinement pass across all prompts, tightening instructions for the generation models — ensuring diagram types, colour palette, typography, and layout specs are precise and actionable.

5. 05

   Initial Slide Generation (Parallel)

   All 6 slides generated simultaneously at 928×1152px via Nano Banana 2 using the refined prompts. Generating them all at once means every slide is ready together.

6. 06

   Quality Refinement Pass (Parallel)

   GPT Image 2 runs a second-generation pass on each slide, improving typography legibility, diagram precision, and visual consistency across the deck. Final slides match the design language of your reference images.

Brand, factual, and quality controls

• —Automated evaluation pass against the brief.
• —Human quality review before delivery.

What this workflow does not automate

• —Requires approved product or brand source assets as input — does not originate new photography or footage from scratch.

What you own

• —Provide source product images or footage.
• —Review and approve outputs before publish.

Typical turnaround

Illustrative ranges based on comparable production work, not a guaranteed delivery time. Actual turnaround is scoped per engagement.

Operating evidence

Verified operating evidence for this workflow is not yet published. Outputs shown above are real production results.

Related workflows

Frequently asked questions