easy-image

Silently translate user's simple descriptions into professional prompts, call image generation APIs, return professional-grade images.

First-time Setup

If ~/.easy-image-skill/config.json not exists, guide user through 4 steps:

1. Select Platform — Jiekou AI(recommended for China) / Novita / PPIO / OpenRouter / WaveSpeed / Google Imagen. Details: references/platforms/*.md
2. API Key — Check ~/.{platform}/config.json, auto-detect existing key or ask user to provide
3. Storage Path — ~/Downloads(default) / ~/Desktop / Custom. This grants blanket download authorization
4. Frequent Scenes (optional) — PPT / Posters / Product Photos / Social Media / Avatar

Save to ~/.easy-image-skill/config.json: {"platform":"jiekou","save_path":"~/Downloads","frequent_scenes":[...]}

Workflow

1. Parse Input

Extract: scene(PPT/poster/product/social media), channel(→auto size, see Channel Mapping below), subject, style, details. If incomplete, ask only what's missing.

2. Match Personal Library

Silently check ~/.easy-image-skill/my-prompts.md for scene+keyword match. No match → use references/templates/{scene}.md.

3. Translate to Professional Prompt

Load template from references/templates/{scene}.md, fill variables, add smart defaults. If image needs text content, explicitly specify language (Chinese input→all text in Simplified Chinese characters, English→all text in English). Terminology: references/glossary.md

4. Select Model

Rules in references/model-selection.md. Summary:

• Default: Gemini 3.1 Flash Image + Grounding (web search ON for any named entity/brand/character)
• High quality: Gemini 3 Pro Image (complex composition + professional photography, ≥2 keyword hits)
• Abstract only: Gemini 3.1 Flash Image without Grounding (pure color/shape descriptions)

5. Show Enhancement Summary

One line before generating: ◇ {template} | +{2-4 key enhancements added}

6. Call API

Platform details: references/platforms/{platform}.md. Hide all technical details from user. Show: ◐ Generating...

7. Save & Display

Auto-download to configured save_path (pre-authorized). Display image immediately, download in background. File naming: {scene}_{brief}_{timestamp}.png

8. Handle Feedback

Satisfied ("good"/"save"/"perfect") → async save to personal library. Adjust request → modify prompt, regenerate. Max 3 adjustment rounds.

Channel Size Mapping

Config Commands

Users can say: "switch to Novita" / "my key is sk-xxx" / "save to desktop" / "show config" / "reset config"

UX Rules

• Auto-detect language (Chinese ratio>0.3 → zh)
• Monochrome status icons: ◇ ◐ ◉ ● (no technical details shown to user)
• Prompts always in English; UI messages follow user language
• Personal library saves are async and non-blocking

Reference Documents