drawthings_skill

Overview

This skill lets you generate images using DrawThings, a local Stable Diffusion implementation on Mac with MLX/CoreML acceleration. It exposes an Automatic1111-compatible API so you can create single images, variations, or batch outputs programmatically. Use the included Python wrapper and scripts to run text-to-image jobs, set presets, and save results with embedded metadata.

How this skill works

The skill calls the DrawThings API (POST /sdapi/v1/txt2img) using parameters like prompt, negative_prompt, steps, sampler_name, cfg_scale, width, height, batch_size, and seed. It runs locally against DRAWTHINGS_URL (default http://127.0.0.1:7860) and saves PNG outputs with generation metadata. Presets and command-line flags make it easy to switch quality, speed, and resolution settings.

When to use it

• Generate images from text prompts for concept art, mockups, or marketing visuals
• Create multiple variations of a character, scene, or product with batch generation
• Experiment with samplers, steps, and CFG scale to tune quality vs. creativity
• Produce reproducible outputs by specifying a seed
• Automate image generation in local workflows without external cloud services

Best practices

• Set DRAWTHINGS_URL to your local DrawThings server before running scripts
• Start with presets (fast/quality/nft) then tweak steps and cfg_scale for final results
• Keep width and height as multiples of 64 to avoid dimension errors
• Use seeds for reproducibility; use -1 for random variations
• Lower steps and use fast samplers for drafts, increase steps and use DPM++ 2M Karras for final images

Example use cases

• Batch-generate 10 character variations for a game or comic using --batch-size
• Produce a high-resolution promo image with --width 1024 --height 1024 --steps 30
• Quickly test multiple samplers and CFG values to find the best visual style
• Save consistent assets by embedding prompt and parameters in output PNG metadata
• Automate nightly generation jobs locally without sending data to external APIs