Image Generation

When to Use

User needs AI-generated images. Agent handles text-to-image, image editing, style transfer, upscaling, and provider selection.

Architecture

User preferences persist in ~/image-generation/. See memory-template.md for setup.

~/image-generation/
├── memory.md      # Current provider, style, projects
└── history.md     # Past generations (optional)

Quick Reference

| Topic | File |

|-------|------|

| Memory setup | memory-template.md |

| Prompt techniques | prompting.md |

| API handling | api-patterns.md |

| OpenAI/DALL-E | openai.md |

| Midjourney | midjourney.md |

| Stable Diffusion | stable-diffusion.md |

| Flux | flux.md |

| Leonardo | leonardo.md |

| Ideogram | ideogram.md |

| Replicate | replicate.md |

Core Rules

1. Check Memory First

Read ~/image-generation/memory.md for user's provider, preferred styles, and project context.

2. Draft Before Final

• -Start at 512x512 or 1024x1024 to validate prompt
• -Generate 4+ variations
• -Only upscale the winner

3. Provider Selection by Task

| Task | Best Provider |

|------|---------------|

| Photorealism | Midjourney, Flux Pro |

| Text in images | Ideogram, DALL-E 3 |

| Fast iteration | Flux Schnell, Leonardo |

| Maximum control | Stable Diffusion |

| Inpainting/editing | DALL-E 3, Stable Diffusion |

| Budget API | Replicate, Leonardo |

4. Prompt Structure

• -Subject first: "A red fox" not "In the forest there is a red fox"
• -Style keywords: "cinematic lighting", "oil painting", "studio photography"
• -Be specific: "golden hour sunlight" not "good lighting"
• -Match aspect ratio to content: 1:1 portraits, 16:9 landscapes

5. Update Memory

| Event | Action |

|-------|--------|

| User chooses provider | Save to memory.md |

| Style works well | Note in memory.md |

| New project started | Add to memory.md |

Common Traps

• -Hands/fingers wrong → regenerate or use inpainting
• -Text garbled → use Ideogram or add text in post-production
• -Faces distorted → add "detailed face" to prompt, use face-fix models
• -Style inconsistent → lock seed, use reference images
• -Watermarks appearing → check model training, use clean models

Security & Privacy

Data that leaves your machine:

• -Prompts sent to chosen AI provider for generation

Data that stays local:

• -Provider preferences in ~/image-generation/
• -No telemetry or analytics

This skill does NOT:

• -Store generated images (provider handles storage)
• -Access files outside ~/image-generation/

External Endpoints

| Provider | Endpoint | Data Sent | Purpose |

|----------|----------|-----------|---------|

| OpenAI | api.openai.com | Prompt text | DALL-E generation |

| Midjourney | discord.com | Prompt text | Image generation |

| Stability AI | api.stability.ai | Prompt text | Stable Diffusion |

| Replicate | api.replicate.com | Prompt text | Flux, SD models |

| Leonardo | cloud.leonardo.ai | Prompt text | Leonardo generation |

| Ideogram | api.ideogram.ai | Prompt text | Text-in-image |

Endpoints depend on chosen provider. No other data is sent externally.

Trust

By using this skill, prompts are sent to third-party AI providers (OpenAI, Midjourney, Stability AI, etc.).

Only install if you trust these services with your prompts.

Feedback

• -If useful: clawhub star image-generation
• -Stay updated: clawhub sync