Metadata

name: smtools-image-generation
description: Generate images from text prompts using AI models via OpenRouter, Kie.ai, or YandexART. Use when the user asks to generate, create, draw, or illustrate an image.
metadata: {"openclaw":{"requires":{"bins":["python3"],"env":["OPENROUTER_API_KEY"]},"primaryEnv":"OPENROUTER_API_KEY","optionalEnv":["KIE_API_KEY"]}}

Image Generation Skill

Generate images from text prompts. Default provider is OpenRouter (synchronous). Alternative: Kie.ai (async, task-based).

When to Activate

Activate when the user asks to:

Run the generation script with an absolute path to avoid directory change prompts:

bash SKILL_DIR/scripts/run.sh --prompt "PROMPT" [OPTIONS]

Replace SKILL_DIR with the absolute path to this skill's root directory.

Flag	Description
`-p, --prompt`	Text prompt (required)
`--provider`	`openrouter` (default), `kie`, or `yandexart`
`-m, --model`	Model name (provider-specific)
`-i, --input`	Input image for editing (path or URL)
`-o, --output`	Output file path
`-c, --config`	Path to config.json
`--list-models`	List available models
`-v, --verbose`	Debug output to stderr

The script outputs JSON to stdout:

{"status": "ok", "image_path": "/absolute/path/to/image.png", "model": "google/gemini-3.1-flash-image-preview", "provider": "openrouter"}

After successful generation, show the user the image path and confirm the image was created.

When the user wants to edit, modify, or transform an existing image, use the -i flag to pass the input image:

bash SKILL_DIR/scripts/run.sh -p "EDITING INSTRUCTION" -i /path/to/source/image.png

How to decide:

User says "draw/generate/create X" → generate from scratch (no -i)
User says "edit/change/modify this image" or references an existing image file → use -i with the path to that image
User provides an image path and an editing instruction → use -i

The editing prompt should describe what to change, e.g. "Add sunglasses", "Make the background blue", "Remove the text".

OpenRouter (default): Fast, synchronous. Models: google/gemini-3.1-flash-image-preview, openai/gpt-image-1, google/imagen-4, stabilityai/stable-diffusion-3. Requires OPENROUTER_API_KEY.
Kie.ai: Async task-based. Models: nano-banana-2, flux-ai, midjourney, google-4o-image, ghibli-ai. Requires KIE_API_KEY. Use when the user explicitly requests Kie.ai or a Kie-specific model.
YandexART: Async task-based. Models: yandex-art/latest. Requires YANDEX_IAM_TOKEN and YANDEX_FOLDER_ID. Use when the user explicitly requests YandexART or Yandex.

Basic generation:

bash SKILL_DIR/scripts/run.sh -p "A serene mountain lake at sunset"

Specific model:

bash SKILL_DIR/scripts/run.sh -p "Cyberpunk cityscape" -m "google/imagen-4"

Kie.ai provider:

bash SKILL_DIR/scripts/run.sh -p "Studio Ghibli forest" --provider kie -m ghibli-ai

Edit an existing image:

bash SKILL_DIR/scripts/run.sh -p "Add a rainbow to the sky" -i /path/to/photo.png

Custom output path:

bash SKILL_DIR/scripts/run.sh -p "A red fox" -o /tmp/fox.png

Error	Action
Missing API key	Tell the user to set the environment variable (`OPENROUTER_API_KEY` or `KIE_API_KEY`)
Network/timeout error	Retry once. If still failing, inform the user
No image in response	Show the raw error from the JSON output
Kie.ai task timeout	Inform user that generation took too long, suggest retrying

If the skill has not been set up yet, run:

bash SKILL_DIR/setup.sh