News-to-Social Pipeline
Every story wants its own social card, and making them by hand does not keep up with a news feed.
The workflow: six feeds merged and scored, headlines rewritten by a local model, and a social image rendered per story
Problem
Every story wants its own social card, with a headline that works at a glance and an image that suits it. Making them by hand does not keep up with a news feed.
Approach
An n8n workflow merges the RSS feeds, scores the stories, and hands each headline to a local language model to tighten. An image service then renders the card from HTML templates: the article's own image when there is one, a background generated with FLUX through ComfyUI when there is not, and a category gradient as the last resort. Colour detection reads each background and picks text and overlay colours that stay legible on it.
Result
Feed-to-card automation with three image fallbacks and a template system that new layouts drop into, running on one GPU machine so no article or prompt leaves it. A Nodevex product.
Stack
n8n for orchestration, Ollama for the headline model, ComfyUI and FLUX for generated backgrounds, a Node.js image service, and Docker Compose to run it as one stack.

