Stable Diffusion prompt engineering, now with actual engineering
A templating language that brings conditionals, recursion, and 70+ shortcodes to your AI image prompts.

What it does
Unprompted is a templating engine that plugs into Automatic1111’s Stable Diffusion WebUI (and ComfyUI) to turn your prompts into programmable scripts. Instead of hand-crafting every prompt, you write [shortcodes] — like [if], [choose], [file] imports, and nested logic — that generate, mutate, and organize your text at runtime. It also exposes WebUI variables (negative_prompt, cfg_scale, etc.) for programmatic read/write.
The interesting bit
The author seems genuinely allergic to the usual “prompt engineering” hand-waving: the whole pitch is that natural-language shortcodes should be readable by non-programmers, yet the feature list includes recursion, custom configs, and a [txt2mask] shortcode that does CLIP-based segmentation. That’s a wider gap than most template engines try to bridge.
Key highlights
- 70+ built-in shortcodes including conditionals, wildcards, file imports, and NLP helpers (auto-pluralization, synonym lookup, verb conjugation)
- Can read and write any WebUI-exposed variable programmatically
- Supports nested shortcodes, recursion, and custom shortcode extensions
- Standalone pip install available for use outside the WebUI
- Also ships as a ComfyUI node
Caveats
- The README’s “Swiss Army knife” and “powerful templating language” claims are self-assessed; no performance or comparison benchmarks are provided
- Premium template packs are advertised mid-documentation, which may or may not signal where sustained development effort is directed
- The “natural language readability” promise is plausible but unverified — your mileage may vary with nested recursion
Verdict Worth a look if you’re generating enough Stable Diffusion images that manual prompting feels like data entry. Skip it if your prompts are one-liners and you like them that way.