Open Image Prompt Library

Open Image Prompt Library

Use the generated catalog at data/prompt-library.json as the source of truth. It contains the same original cases and templates as the website, plus cover values, variables, model compatibility, evaluation status, and limitations.

Select the right asset

1. Start from the requested deliverable, then refine by workflow, capability, and target model.
2. Return 1–3 matching cases with their title, why they match, cover values, variables, and limitations.
3. When the user needs a new brief rather than an example, pair the best case with one compatible industrial template.
4. Preserve each {{variables}} ID exactly when adapting a prompt. Use variable_labels.zh or variable_labels.en to explain its meaning in the user's language; only fill the ID after collecting the user's brief.

Quality rules

• Treat evaluation.status: draft as an unverified starting point; never claim it is tested output.
• Include any listed limitations when they materially affect the request.
• Prefer universal prompts unless the user explicitly names a target model.

Response shape

For a recommendation, use this compact structure:

### 推荐案例
- `case-id` — why it matches; compatible models; verification status.

### 推荐模板
- `template-id` — variables to collect before generating.

### 可直接使用的提示词
<adapted prompt with variables resolved or clearly marked>

Use skills/img2prompt when the user supplies a reference image and asks to reproduce or analyze it. This skill selects reusable library assets; it does not infer an image's visual specification.