Sigil - AI Prompt Translator for Stable Diffusion

Sigil - type English, get SD prompts. A glowing green magical sigil transforms plain text into tag bubbles.

You know what you want to see. Just say it. Type plain English, get Stable Diffusion prompts that actually work. You don't need to memorize tags or read tutorials.

Get Sigil - $7.99 One-time purchase, fully offline.

The Problem

You downloaded Stable Diffusion. You found a cool model. You opened the prompt box and... now what? What do you type? What are "booru tags"? Why does "a pretty girl in a forest" give you garbage?
Sigil fixes this. You type what you want in plain English, and Sigil translates it into the tags that Stable Diffusion needs. You can have your first image in about 15 minutes.

Features

Natural Language Input

Describe what you want to see in plain English. Something like "a knight in silver armor at sunset" and Sigil turns it into the tags SD needs. Custom model trained on 300K+ tag translations with SFW/NSFW toggle.

Tag Validation

Every tag is checked against 234,000+ known Danbooru and e621 tags. Valid tags show up green, unknown ones show red, so you can tell what works before you generate.

Character Detection

Type a character name and Sigil finds them. 265+ characters with alias support and variant picker. One click injects the right appearance tags.

Searchable Tag Database

Search tags right from the input field. Find tags you didn't know existed, see popularity counts, add them to your prompt.

Prompt History

Every translation is auto-saved. Pin favorites, rename entries, reuse prompts, and edit tags. Your work is never lost.

Quality Presets

One-click quality tags for Pony, Illustrious, or generic models. Each model type has its own "magic words" and Sigil helps you pick the right ones.

Smart VRAM Management

Low VRAM mode automatically loads and unloads the AI model around each translation, so Sigil and SD can share the same GPU.

Fully Offline

One-time model download, then everything runs on your machine. No internet needed, no cloud, no API keys.

Which models work best?

Sigil works best with booru-trained models like Pony, Illustrious, and NoobAI. These models were trained on tagged image data, so they respond well to the tags Sigil generates.

For realistic or SDXL-based models, plain English descriptions often work just as well on their own. Sigil can still help, but those models don't rely on booru tags the same way.

Sigil v1.2 includes an experimental custom-trained model (300K+ examples) with SFW/NSFW toggle for better tag accuracy.

From Zero to First Image

No experience needed. 15 minutes, start to finish.

  1. Download ComfyUI Free, open-source image generation software.
  2. Grab a model from CivitAI Pick one that looks cool. PonyXL, NoobAI, and RealisticPony are great starting points.
  3. Drop the model in ComfyUI's models folder That's the only setup ComfyUI needs.
  4. Open Sigil, type what you want "A girl with red hair in a cherry blossom forest." Plain English.
  5. Send to ComfyUI, get your image Done. You just generated your first AI image. Everything else is optional.

Learn at Your Own Pace

Sigil includes short explanations of SD concepts built right in. Not full tutorials, just enough to know what something does and whether you need it.

What are LoRAs?

Extra training that teaches SD specific things. Like mods for a game. Optional.

What is CFG?

How closely SD follows your prompt. Higher = more literal. 7 is a safe default.

What are samplers?

Different ways SD builds the image. Some faster, some prettier. Euler works fine.

Model compatibility

Pony models need Pony LoRAs. SDXL needs SDXL LoRAs. Not everything mixes.

Prompt weights

(tag:1.2) means "more of this." Default is 1.0. Don't go crazy.

Negative prompts

Tags for things you DON'T want. Bad hands, extra fingers. Common ones are built in.

System Requirements

Minimum

  • Windows 10 (64-bit) or Linux (Debian/Ubuntu)
  • 8 GB RAM
  • GPU with 6 GB VRAM, or CPU (slower)
  • 5 GB storage

Recommended

  • 16 GB RAM
  • 8+ GB VRAM
  • NVIDIA / AMD / Intel Arc

Sigil includes a Low VRAM mode for GPUs under 14GB that automatically manages model loading so Sigil and SD can share the GPU.

What's Next

Character detection, prompt history, dual tag database (v1.1), and custom-trained model with SFW/NSFW toggle (v1.2) are live. Here's what's coming:

v2.0

Direct ComfyUI prompt sending. Favorites and quick-access prompts. Improved model accuracy with more training data.

v3.0

Community ecosystem packs: curated model + VAE + LoRA combos that just work together. One-click setup for beginners.

Sigil development roadmap

Ready to try Stable Diffusion?

Skip the tutorials. Just say what you want to see.

Get Sigil - $7.99 One-time purchase, fully offline, no subscriptions.