Recipe: general image

Plain text-to-image via OpenAI/Azure's gpt-image-2. No pixel-art pipeline, no palette quantize — just a PNG.

Command

python3 scripts/generate_image.py \
  --prompt "a misty mountain valley at dawn, oil painting" \
  --size 1024x1024 --quality high \
  --output /tmp/valley.png

Flags explained

Flag	Meaning
--size	Common values include 1024x1024, 1536x1024, 1024x1536, 2048x2048, 2048x1152, 3840x2160, 2160x3840, or auto.
--quality	low / medium / high. Higher quality costs more tokens and takes longer.
--provider	auto, openai, or azure.
--deployment NAME	Override the default gpt-image-2 model/deployment name.
--output	Path to write the PNG. Parent directories are created.

When to use this instead of the pixel-art scripts

• Non-game art — illustrations, concept art, reference imagery.
• Source material to feed into pixelize.py later.
• Any time you need a raw generation without the rembg / palette-quantize / outline pipeline.

Tips

• For custom sizes, use dimensions that are multiples of 16, no more than 3840 px on either edge, within a 3:1 aspect ratio, and between 655,360 and 8,294,400 total pixels.
• Long prompts benefit from explicit style anchors (e.g. oil painting, photograph, anime screencap). The model follows stylistic keywords more reliably than mood words.
• Feed the raw output into the Pixelize existing recipe to convert it to pixel art after the fact.