saa-agent

What This Skill Does

The saa-agent skill acts as a robust bridge between the OpenClaw AI agent and the Character Select Stand Alone App (SAA) image generation backend. By leveraging the command-line interface of the SAA tool, this skill empowers agents to generate high-quality AI images directly from terminal commands. It supports advanced features such as regional prompting, model selection, and fine-tuning parameters like CFG scale and sampling steps, making it an essential tool for creators who prefer local control over their generation pipeline.

Installation

To integrate this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/mirabarukaso/saa-agent. Before the tool can be used, ensure the SAA backend (version 2.4.0 or higher) is running locally or on a reachable server. You must have the SAAC (SAA Client) feature enabled and a valid WebSocket address to establish communication. Mac users should verify if their environment requires python3 instead of python to execute the agent script correctly.

Use Cases

This skill is ideal for AI-assisted creative workflows, such as character design, concept art development, and rapid prototyping. It is particularly useful when users need to switch between different checkpoint models (like Illustrious SDXL) without leaving their current environment. Regional prompting allows for precise control over split-screen composition, enabling the agent to place distinct subjects on the left and right sides of the image canvas independently, ensuring better structural composition than standard prompt mixing.

Example Prompts

1. "Use the saa-agent to generate an image of a cybernetic knight in a futuristic city using the waiIllustriousSDXL_v160 model at the local websocket address ws://localhost:7860."
2. "I need a split-composition image using regional prompting: put a medieval warrior on the left and a mysterious mage on the right. Please use the SAA tool."
3. "Generate a portrait of a forest elf using the saa-agent. Keep the style consistent with the current model, set steps to 30, and make it 1024x1024."

Tips & Limitations

Always check if the backend is busy before triggering a new request; if you encounter a busy error, do not force an immediate retry. High-resolution upscaling (HiResFix) is a resource-intensive task that should only be enabled if the user explicitly demands it. Remember that the skill requires specific input parameters like the WebSocket address and model name to function. If you are unsure of the current configuration, running python saa-agent.py --help will provide a comprehensive list of all available commands and default values for sampling algorithms and schedulers.