gpt-image

name	gpt-image
description	Generate and edit images with OpenAI gpt-image-2 for PraisonAI PPT slide assets. Use when creating diagrams, charts, or visuals for YAML decks (slide_type image), or any "generate/create/edit image" request for this repo.
disable-model-invocation	true

GPT Image (PraisonAI PPT)

Generate and edit images via gpt-image-2 CLI scripts in this repo. Pair with ppt-yaml-deck-workflow to embed results in decks.

Skill root: .cursor/skills/gpt-image/

Setup

Scripts use PEP 723 metadata; run with uv run (installs openai on first use).

OPENAI_API_KEY from:

1. Shell export (export OPENAI_API_KEY=sk-...) — preferred
2. .cursor/skills/gpt-image/.env (copy from .env.example)

Scripts

Script	Use
scripts/generate.py	Text → image (no input file)
scripts/edit.py	Text + 1–10 images → image
scripts/validate_skill.py	Local checks (no API); optional --generate smoke test

Quick decision

• No input images → generate.py
• Has input image(s) → edit.py
• Slide asset for YAML → generate at 1536x864 (widescreen), save under assets/generated/

Generate (slide diagram example)

cd /Users/praison/ppt-package

uv run .cursor/skills/gpt-image/scripts/generate.py \
  --prompt "Simple sermon diagram: two columns 'My Power' vs 'His Power', dark blue background, gold and green text labels only, no logos, no watermark, clean flat design for a PowerPoint slide." \
  --size 1536x864 \
  --quality medium \
  --output assets/generated/power_diagram.png

Use in YAML deck

- slide_type: image
  image_path: assets/generated/power_diagram.png
  image_fit: contain
  reference: 100% His Power
  text: Trust grace, not self-effort

Then regenerate the deck:

python3 -m praisonaippt.cli -i examples/how_to_prevent_delay.yaml -o examples/how_to_prevent_delay.pptx

generate.py flags

Flag	Default	Notes
--prompt	required
--output	output.png	Use repo paths under assets/generated/
--size	1024x1024	Slides: 1536x864 (widescreen)
--quality	medium	high for dense text
--output-format	png
--n	1	1–4 variants

edit.py

uv run .cursor/skills/gpt-image/scripts/edit.py \
  --prompt "Instruction" \
  --images input.png \
  --output assets/generated/edited.png \
  --size 1536x864

Validate skill (no API cost)

uv run .cursor/skills/gpt-image/scripts/validate_skill.py

Validate with live generation

uv run .cursor/skills/gpt-image/scripts/validate_skill.py --generate

Expect ✓ Saved: assets/generated/skill_test.png and a non-empty PNG.

Prompting

Read references/prompting-guide.md before writing prompts. For slide diagrams: flat design, high contrast, no logos/trademarks/watermarks, legible text in quotes.

Size reference (gpt-image-2)

Use	Size
Widescreen slide	1536x864
Portrait slide	1024x1536
Square	1024x1024

Edges must be multiples of 16; ratio ≤ 3:1.

Errors

Scripts print clear messages for missing API key, invalid size, or missing input files.