AI Image Generation
📖 技能介绍
Setup
On first use, read setup.md.
When to Use
更多技能请访问小葱技能站7w4.net。
User needs AI-generated visuals, edits, or consistent image sets. Use this skill to pick the right model, write stronger prompts, and avoid outdated model choices.
Architecture
User preferences persist in ~/image-generation/. See memory-template.md for setup.
~/image-generation/
├── memory.md # Preferred providers, project context, winning recipes
└── history.md # Optional generation logQuick Reference
| Topic | File |
|---|---|
| Initial setup | setup.md |
| Memory template | memory-template.md |
| Migration guide | migration.md |
| Benchmark snapshots | benchmarks-2026.md |
| Prompt techniques | prompting.md |
| API handling | api-patterns.md |
| GPT Image (OpenAI) | gpt-image.md |
| Gemini and Imagen (Google) | gemini.md |
| FLUX (Black Forest Labs) | flux.md |
| Midjourney | midjourney.md |
| Leonardo | leonardo.md |
| Ideogram | ideogram.md |
| Replicate | replicate.md |
| Stable Diffusion | stable-diffusion.md |
Core Rules
1. Resolve aliases to official model IDs first
Community names shift quickly. Before calling an API, map the nickname to the provider model ID.
| Community label | Official model ID to try first | Notes |
|---|---|---|
| Nano Banana | gemini-2.5-flash-image-preview |
Common nickname, not an official Google model ID |
| Nano Banana 2 / Pro | Verify provider docs | Usually a provider preset over Gemini image models |
| GPT Image 1.5 | gpt-image-1.5 |
Current OpenAI high-tier image model |
| GPT Image mini / iMini | gpt-image-1-mini |
Budget/faster OpenAI variant |
| FLUX 2 Pro / Max | flux-pro / flux-ultra |
Many platforms rename these SKUs |
2. Pick models by task, not by hype
| Task | First choice | Backup |
|---|---|---|
| Exact text in image | gpt-image-1.5 |
Ideogram |
| Multi-turn edits | gemini-2.5-flash-image-preview |
flux-kontext-pro |
| Photoreal hero shots | imagen-4.0-ultra-generate-001 |
flux-ultra |
| Fast low-cost drafts | gpt-image-1-mini |
imagen-4.0-fast-generate-001 |
| Character/product consistency | flux-kontext-max |
gpt-image-1.5 with references |
| Local no-API workflows | flux-schnell |
SDXL |
3. Use benchmark tables as dated snapshots
Benchmarks drift weekly. Use benchmarks-2026.md as a starting point, then recheck current rankings when quality is critical.
4. Draft cheap, finish expensive
Start with 1-4 low-cost drafts, pick one, then upscale or rerender only the winner.
5. Keep a fallback chain
If the preferred model is unavailable, fallback by tier: 1) same provider lower tier, 2) cross-provider equivalent, 3) local/open model.
6. Treat DALL-E as legacy
OpenAI lists DALL-E 2/3 as legacy. Do not use them as default for new projects.
Common Traps
- Using vendor nicknames as model IDs -> API errors and wasted retries
- Assuming "Nano Banana Pro" or "FLUX 2" are universal IDs -> provider mismatch
- Copying old DALL-E prompt habits -> weaker output vs modern GPT/Gemini image models
- Comparing text-to-image and image-editing scores as if they were the same benchmark
- Optimizing every draft at max quality -> cost spikes without quality gain
Security & Privacy
Data that leaves your machine:
- Prompt text
- Reference images when editing or style matching
Data that stays local:
- Provider preferences in
~/image-generation/memory.md - Optional local history file
This skill does NOT:
- Store API keys
- Upload files outside chosen provider requests
- Persist generated images unless user asks to save them
External Endpoints
| Provider | Endpoint | Data Sent | Purpose |
|---|---|---|---|
| OpenAI | api.openai.com |
Prompt text, optional input images | GPT Image generation/editing |
| Google Gemini API | generativelanguage.googleapis.com |
Prompt text, optional input images | Gemini image generation/editing |
| Google Vertex AI | aiplatform.googleapis.com |
Prompt text, optional input images | Imagen 4 generation |
| Black Forest Labs | api.bfl.ai |
Prompt text, optional input images | FLUX generation/editing |
| Replicate | api.replicate.com |
Prompt text, optional input images | Hosted third-party image models |
| Midjourney | discord.com |
Prompt text | Midjourney generation via Discord workflows |
| Leonardo | cloud.leonardo.ai |
Prompt text, optional input images | Leonardo generation/editing |
| Ideogram | api.ideogram.ai |
Prompt text | Typography-focused image generation |
No other data is sent externally.
Migration
If upgrading from a previous version, read migration.md before updating local memory structure.
Trust
This skill may send prompts and reference images to third-party AI providers. Only install if you trust those providers with your content.
Related Skills
Install with clawhub install <slug> if user confirms:
image-edit- Specialized inpainting, outpainting, and mask workflowsvideo-generation- Convert image concepts into video pipelinescolors- Build palettes for visual consistency across assetsffmpeg- Post-process image sequences and exports
Feedback
- If useful:
clawhub star image-generation - Stay updated:
clawhub sync
🤖 AI 评测
这个 Skill 质量不错，结构清晰、指引详细，对主流 AI 图像生成工具的使用很有帮助。优点是内容覆盖全面、框架完整、参考价值高；不足是暂时只有英文版本，缺乏多语言支持，且缺少直观的图像案例供参考。
