← Nexus Automation

Brand reimaginer

Your brand. A fresh carousel. Every time.

Start with your website and a topic. Get five coordinated slides using your real colours, typography and logo, ready to publish.

5 slides per run. About $0.20 in AI generation cost.

Generated hook slideMatching feature slide
Two slides from one real carousel

The project

Brand Reimaginer Automation

The problem

Agencies and freelancers burn hours per client manually matching brand colors, fonts, logos, and product photography for every social carousel. And AI-generated text and logos hallucinate almost every time, which makes that shortcut unusable for real client work.

The build

A 24-node n8n pipeline takes one URL and reads the brand's real computed styles via a headless browser, not regex-scraped HTML, so it works identically on modern React/Next.js and Tailwind/CSS-in-JS sites where colors never appear as literal text. An LLM derives a design system from that real data, Nano Banana 2 generates on-brand art conditioned on the brand's own product photography, and all typography plus the real logo file are composited in a separate programmatic layer, completely outside the AI's control, which is what eliminates hallucinated text and distorted logos entirely. Multi-site runs process one site fully at a time, with full item context re-attached at every async hop, closing a real class of data-crossover bug found during build.

The outcome

5 on-brand slides per run, zero manual design steps, ~$0.20 AI generation cost per full 5-slide carousel, fully automated from URL submission to finished PNGs.

5On-brand slides per run
0Manual design steps
$0.20AI cost per 5-slide carousel
24Orchestration nodes

How it runs

  1. Trigger: form intake

    A single form submission, just a URL and a topic, kicks off the entire run. No manual brief, no back-and-forth.

  2. Extract: read the real brand

    A headless browser reads the site's final computed styles and real product photography, not guesses from a template or regex-scraped HTML.

  3. Design system: derived, not templated

    An LLM turns the extracted brand data into a structured design system (color roles, font pairing, layout rules) locked to the brand's exact hex values.

  4. Generate: reimagine, don't invent

    Nano Banana 2 generates background art conditioned on the brand's own reference photos, fed in as real images. A reimagining of a real asset, never an invented one.

  5. Composite: type & logo, never AI-rendered

    Typography and the brand's real logo file are rendered programmatically in a separate layer, completely outside the AI's control. That's the reason nothing is ever hallucinated.

Proof, not promises

Real output — the hook slide: on-brand art conditioned on the brand's own product photography.
Real output — the hook slide: on-brand art conditioned on the brand's own product photography.
The feature slide from the same run, locked to the brand's exact hex values.
The feature slide from the same run, locked to the brand's exact hex values.
An explainer slide: typography rendered programmatically, never AI-drawn.
An explainer slide: typography rendered programmatically, never AI-drawn.
The client-quote slide — text composited in a separate layer, outside the AI's control.
The client-quote slide — text composited in a separate layer, outside the AI's control.
The closing slide of the 5-slide run: the brand's real logo file, never hallucinated.
The closing slide of the 5-slide run: the brand's real logo file, never hallucinated.

In short

New client, new brand, same 5 slides, zero manual design, every time.

n8nOpenRouter (GPT-4o)Nano Banana 2PuppeteerRemotionTypeScript
Have a process worth automating? →