@markus-global/cli 0.8.2 → 0.8.4-rc.0

package/templates/skills/image-generation/SKILL.md

@@ -1,183 +0,0 @@
----
-name: image-generation
-description: Generate images using AI models across multiple providers
----
-
-# Image Generation
-
-You have access to AI image generation tools (prefixed `image-generation__`) that can create
-images from text prompts using multiple providers and models.
-
-## When to use these tools
-
-Use image generation tools when the user wants to:
-- Create, draw, or generate images from a text description
-- Create illustrations, diagrams, concept art, or any visual content
-- Edit or modify an existing image
-- Generate variations of an image
-
-Do NOT use these tools for:
-- Taking screenshots (use Chrome DevTools instead)
-- Converting documents to images (use MarkItDown or other tools)
-- Text-only diagrams where Mermaid/ASCII would suffice
-
-## Available tools
-
-| Tool | Purpose |
-|------|---------|
-| `image-generation__generate_image` | Generate an image from a text prompt |
-| `image-generation__list_providers` | Check which providers are configured and available |
-| `image-generation__edit_image` | Edit/modify an existing image (OpenAI, Stability, Ideogram) |
-
-## Supported providers
-
-The skill auto-detects available providers based on environment variables.
-If no provider is specified, the first available one is used.
-
-| Provider | Env Variable | Default Model | Strengths |
-|----------|-------------|---------------|-----------|
-| OpenAI | `OPENAI_API_KEY` | dall-e-3 | High quality, good prompt following, style control |
-| Azure OpenAI | `AZURE_OPENAI_API_KEY` + `AZURE_OPENAI_ENDPOINT` | dall-e-3 | Enterprise, same quality as OpenAI |
-| Stability AI | `STABILITY_API_KEY` | sd3-large | Fine control, negative prompts, inpainting |
-| Google Imagen | `GOOGLE_API_KEY` | imagen-3.0-generate-002 | Photorealistic, good text rendering |
-| Replicate | `REPLICATE_API_TOKEN` | black-forest-labs/flux-1.1-pro | Wide model selection, Flux models |
-| Tongyi Wanxiang | `DASHSCOPE_API_KEY` | wanx2.1-t2i-turbo | Chinese text support, fast turbo mode |
-| Zhipu CogView | `ZHIPU_API_KEY` | cogview-4 | Chinese text support, multiple styles |
-| SiliconFlow | `SILICONFLOW_API_KEY` | FLUX.1-schnell | Affordable, hosts FLUX & SD & Qwen models, fast |
-| Together AI | `TOGETHER_API_KEY` | FLUX.1.1-pro | High-quality FLUX, competitive pricing |
-| FAL | `FAL_KEY` | flux-pro/v1.1 | Fast inference, Flux Pro/Dev/Schnell |
-| Ideogram | `IDEOGRAM_API_KEY` | V_2 | Excellent text-in-image, supports editing |
-| Baidu ERNIE ViLG | `BAIDU_API_KEY` + `BAIDU_SECRET_KEY` | sd_xl | Chinese ecosystem, Baidu Cloud |
-| Tencent Hunyuan | `HUNYUAN_API_KEY` | hunyuan-image | Chinese ecosystem, Tencent Cloud |
-| Volcengine (Doubao) | `VOLCENGINE_API_KEY` | general_v2.1_L | Chinese ecosystem, ByteDance Cloud |
-
-## Key parameters
-
-### Common parameters (all providers)
-
-- **prompt** (required): Text description of the desired image. Be specific and detailed.
-- **size**: Image dimensions (e.g. `1024x1024`, `1792x1024`). Check provider capabilities for supported sizes.
-- **n**: Number of images to generate (default: 1).
-- **output_dir**: Where to save generated files (default: `~/.markus/generated-images/`).
-- **output_format**: `png`, `jpeg`, or `webp` (default: `png`).
-
-### Provider/model selection
-
-- **provider**: Force a specific provider (e.g. `openai`, `stability`, `zhipu`, `siliconflow`, `fal`). Omit to auto-select.
-- **model**: Specific model name. Each provider has a default, but you can override:
-  - OpenAI: `dall-e-3`, `dall-e-2`, `gpt-image-1`
-  - Stability: `sd3-large`, `sd3-large-turbo`, `stable-image-ultra`, `stable-image-core`
-  - Google: `imagen-3.0-generate-002`
-  - Replicate: `black-forest-labs/flux-1.1-pro`, `black-forest-labs/flux-schnell`
-  - Tongyi: `wanx2.1-t2i-turbo`, `wanx2.1-t2i-plus`
-  - Zhipu: `cogview-4`, `cogview-4-250304`, `cogview-3-plus`
-  - SiliconFlow: `black-forest-labs/FLUX.1-schnell`, `black-forest-labs/FLUX.1-dev`, `black-forest-labs/FLUX.1-pro`, `black-forest-labs/FLUX.1.1-pro`, `stabilityai/stable-diffusion-3-5-large`, `stabilityai/stable-diffusion-3-5-medium`, `Qwen/Qwen-Image`, `deepseek-ai/Janus-Pro-7B`
-  - Together: `black-forest-labs/FLUX.1.1-pro`, `black-forest-labs/FLUX.1-schnell`
-  - FAL: `fal-ai/flux-pro/v1.1`, `fal-ai/flux/schnell`, `fal-ai/flux/dev`
-  - Ideogram: `V_2`, `V_2_TURBO`, `V_1`
-  - Baidu: `sd_xl`
-  - Hunyuan: `hunyuan-image`, `hunyuan-image-fast`
-  - Volcengine: `general_v2.1_L`, `general_v2.0_L`
-
-### Style and quality
-
-- **quality**: `standard` or `hd` (OpenAI, Zhipu)
-- **style**: `natural` or `vivid` (OpenAI), or style-specific values (Tongyi)
-- **negative_prompt**: Describe what to avoid (Stability, Google, Replicate, Tongyi, SiliconFlow, Together, FAL, Ideogram, Baidu, Hunyuan)
-- **seed**: Integer for reproducible results (Stability, Google, Replicate, Tongyi, SiliconFlow, FAL, Ideogram, Baidu, Volcengine)
-
-## Best practices
-
-1. **Check availability first**: If unsure which providers are configured, call
-   `list_providers` before generating. This avoids confusing error messages.
-
-2. **Write detailed prompts**: More specific prompts produce better results. Include subject,
-   style, mood, lighting, composition, and medium (e.g. "oil painting", "photograph",
-   "3D render").
-
-3. **Confirm with user**: Before generating, confirm the prompt and parameters with the user,
-   especially if they gave a vague request. Share the exact prompt you plan to use.
-
-4. **Choose the right provider**: Match provider to use case:
-   - Photorealistic images → OpenAI DALL-E 3, Google Imagen
-   - Artistic/creative control → Stability AI, Ideogram (supports negative prompts, seeds)
-   - Chinese text/content → Tongyi Wanxiang, Zhipu CogView, Baidu ERNIE, Hunyuan, Volcengine
-   - Cutting-edge Flux models → Replicate, FAL, Together AI, SiliconFlow
-   - Budget-friendly → SiliconFlow, Together AI
-   - Text-in-image quality → Ideogram
-   - Image editing/inpainting → OpenAI (DALL-E 2), Stability AI, Ideogram
-
-5. **Report results clearly**: After generation, tell the user:
-   - The file path where the image was saved
-   - Which provider and model were used
-   - The revised prompt (if the provider modified it, e.g. OpenAI)
-
-6. **Handle errors gracefully**: If a provider fails, suggest the user:
-   - Check that the API key is valid and has quota
-   - Try a different provider
-   - Simplify the prompt if content policy rejected it
-
-## Common workflows
-
-### Basic image generation
-```
-1. list_providers → check what's available
-2. generate_image → prompt="A serene mountain landscape at sunset, oil painting style",
-                    size="1792x1024"
-3. Report file path and provider used to user
-```
-
-### Specific provider and model
-```
-1. generate_image → prompt="...", provider="stability", model="sd3-large",
-                    negative_prompt="blurry, low quality", seed=42
-```
-
-### Image editing
-```
-1. edit_image → image_path="/path/to/original.png",
-               prompt="Replace the sky with a starry night",
-               provider="stability"
-```
-
-### Batch generation for comparison
-```
-1. generate_image → prompt="...", provider="openai"
-2. generate_image → prompt="...", provider="stability"
-3. Present both results to user for comparison
-```
-
-## User setup guide
-
-If no providers are configured, guide the user to set environment variables:
-
-```bash
-# --- Global providers ---
-export OPENAI_API_KEY="sk-..."
-export STABILITY_API_KEY="sk-..."
-export GOOGLE_API_KEY="..."
-export REPLICATE_API_TOKEN="r8_..."
-
-# --- Flux / SD hosting platforms ---
-export SILICONFLOW_API_KEY="sk-..."
-export TOGETHER_API_KEY="..."
-export FAL_KEY="..."
-
-# --- Specialized ---
-export IDEOGRAM_API_KEY="..."
-
-# --- Chinese providers ---
-export DASHSCOPE_API_KEY="sk-..."          # Tongyi Wanxiang (Aliyun)
-export ZHIPU_API_KEY="..."                 # Zhipu CogView
-export HUNYUAN_API_KEY="..."               # Tencent Hunyuan
-export VOLCENGINE_API_KEY="..."            # Volcengine (Doubao / ByteDance)
-
-# --- Requires two keys ---
-export AZURE_OPENAI_API_KEY="..."
-export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com"
-
-export BAIDU_API_KEY="..."                 # Baidu ERNIE ViLG
-export BAIDU_SECRET_KEY="..."
-```
-
-These can be set in the shell profile, `.env` file, or Markus settings.