awesome-content-graphics

# Content Graphics One picture for a post, chosen out of a set the user looked at. **The agent running this skill is what draws them.** Not a service, not an MCP server, not a generator installed on the machine: the model reads the material, decides what each picture is, and composes it itself — one at a time, from nothing, in whatever markup it can emit. A rasterizer turns that markup into a PNG; a rasterizer is not a generator and it invents nothing. Nothing here is a template either. There is no layout catalog, no palette pool and no geometry gate — the previous version had all three, and a set built from them was the catalog looked at a hundred times. Every picture in a round is composed for that round. Nothing is fetched, nothing is uploaded, no key is needed and none is looked for. Bundled file (load on demand): - `references/generation.md` — turning a reference into ideas that differ, composing a hundred that stay different, what a second round has to change, and what to check in what was drawn. Read it before the ideas are fixed, and again before a second round is built. ## The five steps ``` 1. Take the reference → text, images, a URL, or several; read for the subject 2. Ask the count → 10 · 20 · 50 · 100 3. Generate and open → the whole set in one folder, opened on the machine 4. Take the answer → a number or a name, or another round on different ideas 5. Save the pick apart → its own folder, opened, and the run sto...