p-image

p-image is a skill for generating and editing images with Pruna's optimized P-Image model family through the inference.sh command-line tool (`belt`). Pruna optimizes popular diffusion models for faster, cheaper inference without sacrificing quality, and this skill wraps that catalogue so an agent can produce images from a single shell command. It solves the practical problem of turning a text prompt or an input image into a finished visual asset without wiring up an SDK, GPU, or model weights locally.

The skill documents four core models: P-Image for fast text-to-image across multiple aspect ratios, P-Image-LoRA for preset artistic styles (photo-realism, pixel-art, cinematic, neon-punk, anime, and others), P-Image-Edit for image editing and multi-image compositing, and P-Image-Edit-LoRA for stylized edits. Every operation is invoked as `belt app run pruna/<model> --input '{...}'` with a JSON payload carrying the prompt, aspect ratio (1:1, 16:9, 9:16, custom, and more), LoRA preset, or source image URLs. It also points to other Pruna-optimized options such as FLUX Dev, FLUX Klein 4B, Qwen Image, and Z-Image Turbo, and to related inference.sh skills for video and text-to-speech.

It targets developers, designers, and content creators who want fast, low-cost AI image generation from a terminal or an agent workflow — product photography mockups, stylized portraits, background edits, collages, and rapid concept iteration. Because it depends entirely on the hosted inference.sh service and the `belt` CLI, it is essentially a documented front-end to that platform rather than a self-contained image engine.