Guides

Images, video & plugins

Local image and video generation, plus a plugin for extracting data into JSON.

Images & video

Install the stable-diffusion.cpp engine (built on the same ggml as llama.cpp) from Settings into its own directory. Start a diffusion model from the Models tab, then in “Images & video” enter a prompt (optionally negative), pick aspect ratio and resolution, and use Advanced for steps, CFG and seed. Some models need extra files: VAE, text encoder, CLIP-L/G or T5-XXL. Video (e.g. Wan2.1) runs on the same server as a queued job you can cancel.

JSON Extractor

Paste text, define a schema (with the builder or raw JSON — or ask for one to be suggested from the text) and the local model returns data in exactly that shape. llama.cpp guarantees valid JSON. Fields not listed in “required” may be skipped by the model.