image-generationlisted

# image-generation How to use the `mcp__openrouter-image__generate_image` tool well. This skill assumes the `openrouter-image-mcp` MCP server is configured and connected. The tool exposes one operation, `generate_image`, that fans out to six different image-gen models on OpenRouter. --- ## 1. Pick the model Decide by use case, not by "which is best" — there is no universal best. | Use case | Model | Why | |---|---|---| | Hero/feature for an article (default) | `nb2` | Balanced; ~$0.07; good text-in-image; fast | | Need 2 variations in one shot | `pro` | Returns 2 images per call; best for complex scenes | | Throwaway draft to compare directions | `flash` or `gpt-mini` | ~$0.04 each; iterate cheap | | Publication-grade photoreal final | `gpt-pro` | Highest fidelity; ~$0.22 | | "ChatGPT-look" stylised art | `gpt` | Distinctive polished aesthetic | | Clean text rendered IN the image | `nb2` / `pro` | Google's text engines beat OpenAI's | | Diagram / wireframe with legible labels | `pro` | Best at composing readable typography | Default to `nb2` unless the user explicitly asks for higher quality or lower cost. Switching to `gpt-pro` triples cost — only do it when it matters. ## 2. Pick the aspect ratio Use `aspect_ratio` parameter — chat-completions has no structured field so the tool prepends it to the prompt. Always set it explicitly; models don't auto-detect intent. | Context | aspect_ratio | |---|---| | Article hero (most blog layouts) | `16:9` | | Square social pos