generate-image

Image Generation — fal.ai (GPT-Image 2)

Generate images by running the bundled Python script via uv run. The script loads the FAL_KEY from fnox internally.

Usage

uv run ./scripts/generate.py --prompt "PROMPT" [OPTIONS]

Options

Flag	Default	Choices
--prompt	(required)	Free text, 2-32k chars
--model	gpt-2	gpt-2, gpt
--size	1024x1024	1024x1024, 1536x1024, 1024x1536, 1:1, 16:9, 9:16, 4:3, 3:4, 21:9
--quality	medium	low, medium, high
--num-images	1	1–4
--format	png	jpeg, png, webp
--background	auto	auto, transparent, opaque (--model gpt only)
--filename	(none)	Output base name without extension (uses --output-dir)
--output	(none)	Output path stem without extension (overrides --output-dir)
--output-dir	OS temp dir	Any directory path

Model Selection

Model	Endpoint	Best for
gpt-2	openai/gpt-image-2	Default — current OpenAI image model
gpt	fal-ai/gpt-image-1.5	Transparent backgrounds (gpt-image-2 cannot produce true alpha PNGs)

Auto-select gpt only when the user requests a transparent or non-opaque background. Otherwise stick with the default gpt-2.

Workflow

1. Parse the user's request into a prompt and optional parameters.
2. Use gpt-2 by default. Switch to gpt only if transparency is needed.
3. Run the script — it prints the saved file path(s) to stdout, one per line.
4. Display each saved image to the user using the Read tool.
5. Tell the user where the file(s) were saved.