genimg
genimg — Image Generation
Text-to-image and image-to-image via any OpenAI-compatible image API.
Usage
/genimg <prompt> # text-to-image
/genimg <image_path> <prompt> # image-to-image
/genimg init # interactive setup wizard
Setup
If ~/.genimg/config.json doesn't exist, the script will print a friendly error with the expected format and suggest running /genimg init.
/genimg init — Interactive Setup
When the user runs /genimg init, walk them through creating or updating ~/.genimg/config.json:
- Ask for provider name (e.g. "volcengine", "openai", "together")
- Ask for base URL (the OpenAI-compatible API endpoint, e.g.
https://ark.cn-beijing.volces.com/api/v3) - Ask for API key
- Ask for model name (e.g.
doubao-seedream-3-0-t2i-250415,dall-e-3) - Ask for default size (e.g.
2048x2048,1024x1024) - Write config to
~/.genimg/config.json, setting this asdefault_provider
If a config already exists, show current providers and offer to add a new one or update an existing one.
Config
Provider settings live in ~/.genimg/config.json. Structure:
{
"providers": {
"<name>": {
"base_url": "...",
"api_key": "...",
"model": "...",
"default_size": "2048x2048"
}
},
"default_provider": "<name>"
}
default_size must be valid for the configured provider and model. Do not assume 1024x1024 is universally accepted. For the current volcengine configuration, use 2048x2048.
To add a new provider, just add an entry to providers. To switch default, change default_provider.
Running
python3 <skill-path>/scripts/generate.py \
--prompt "<prompt>" \
--output "<output_path>" \
[--provider "<provider_name>"] \
[--image "<input_image_path>"] \
[--size "2048x2048"]
--provider: optional, defaults todefault_providerin config--image: enables img2img mode, size defaults toauto--size: override the provider's default size
Workflow
- Run
scripts/generate.pywith prompt and output path - Pick a short descriptive filename (English, kebab-case)
- Show the result to the user with the Read tool
More from verneagent/tiny-skills
inscribe
Capture rules, conventions, or code style guidelines into documentation files. Use when the user says "inscribe", "learn", "remember this rule", "add convention", or wants to persist coding guidelines.
4lark-share
Share a knowledge insight or message with a team via Lark group chat webhook. Use when the user says "share", "lark-share", or wants to send a formatted message to a Lark group.
4multi-gh
Fix and standardize GitHub multi-account workflows with gh account switching, SSH host aliases, and safe remote setup. Use when creating repos, pushing code, or diagnosing GitHub auth mismatches across multiple identities.
2retro
Retrospective on mistakes or new conventions — analyze patterns, find root causes, and propose deterministic prevention (static checks > lint > tests > runtime > review > docs). Use when the user says "retro", "反省", "复盘", "怎么防", "how to prevent", or wants to enforce a new convention.
2wksp
Open a new iTerm2 tab with Claude in a worktree or folder, optionally entering handoff mode.
1netmap
Show all network interfaces, LAN neighbors, and Tailscale peers on the current machine. Use when the user asks about network status, neighbors, connected devices, or Tailscale peers.
1