imagegen

A Pi skill for making and editing images with GPT image generation models.

Use it when you want a real bitmap image: a hero image, product shot, mockup, texture, sprite, cutout, or an edit to an existing picture.

The skill keeps the everyday instructions short. The longer notes live in references/ so the agent only reads them when something actually needs them.

Install

This skill requires the pi-better-skills extension for compatibility:

pi install git:https://github.com/edxeth/pi-better-skills

Then install the skill:

npx skills add edxeth/imagegen

To repair or upgrade an existing Pi installation non-interactively:

npx skills add edxeth/imagegen \
  --skill imagegen \
  --agent pi \
  --global \
  --yes

The repository keeps SKILL.md, scripts/, and references/ together under imagegen/. This ensures the skills CLI installs the complete bundle rather than only SKILL.md.

Setup

Add this to your shell config, for example ~/.zshrc:

export PI_IMAGEGEN_BASE_URL='http://localhost:8317/v1'
export PI_IMAGEGEN_API_KEY='dummy'
export PI_IMAGEGEN_MODEL='gpt-image-2'
export PI_IMAGEGEN_OUTPUT_DIR="$HOME/.pi/generated_images"

Then reload your shell:

source ~/.zshrc

Quick start

From the repository root:

python imagegen/scripts/image_gen.py generate \
  --prompt "A tiny red cube on a plain white background, no text"

Save into a project:

python imagegen/scripts/image_gen.py generate \
  --prompt "A clean product photo of a matte ceramic coffee mug, no text" \
  --out public/images/mug-hero.png

Edit an image:

python imagegen/scripts/image_gen.py edit \
  --image input.png \
  --prompt "Replace only the background with warm sunset light; keep the subject unchanged"

From an installed skill directory, the equivalent path remains scripts/image_gen.py.