Stable Diffusion Image Generation API

Problem

Stable Diffusion workflows often require manually starting WebUI, copying prompts, tuning parameters, and post-processing images, which makes the process hard to reuse in agent or script pipelines. Stable Diffusion Image Generation API wraps the local SD WebUI --api surface into a command-line interface, allowing automated flows from status checks and model listing to image generation.

How It Works

The skill centers on sd_client.py. It first checks whether http://127.0.0.1:7860 or SD_WEBUI_URL is reachable via status, then inspects available checkpoints with list-models. Its core capabilities include:
- Text-to-image: txt2img handles positive prompts, negative prompts, --steps, --cfg-scale, sampler, dimensions, and seed.
- Image-to-image: img2img uses denoising-strength to control how much of the source image is preserved.
- Inpainting: inpaint with mask-fill modes edits specific regions.
- ControlNet: supports modules such as openpose, canny, depth, and lineart for structural control.
- Upscaling and model management: invokes algorithms such as R-ESRGAN or SwinIR, switches checkpoints, or applies LoRA syntax.
prompt_helper.py maps Chinese descriptions to common Stable Diffusion English style terms, such as photorealistic, anime, oil painting, cyberpunk, and traditional Chinese art.

Boundaries

The skill does not bundle Stable Diffusion model code and does not deploy WebUI itself; a local or remote AUTOMATIC1111 SD WebUI must already be running with --api enabled. Output quality depends on the model, VRAM, resolution, --steps, sampler, and ControlNet weights. Remote usage also requires attention to timeout, cross-origin access, and port reachability.