Bringing Booru tag discipline to Stable Diffusion prompts

What it does

Tag Autocomplete is a browser-side extension for the AUTOMATIC1111 Stable Diffusion WebUI that pops up completion hints while you type prompts. It draws from Danbooru and e621 tag databases, plus your local wildcards, embeddings, LoRAs, and custom “chant” presets, to suggest the exact tokens your model likely recognizes. The goal is to replace fuzzy prompt guesswork with the vocabulary the model actually learned.

The interesting bit

The extension treats prompt engineering like code completion: it offers live previews of Extra Networks, filters embeddings by SD 1.x versus 2.x compatibility, and can even append known LoRA trigger words automatically. That last trick leans on either the native WebUI v1.5.0+ keyword system or the third-party model-keyword extension, which means it tries to solve the “what magic phrase activates this model?” problem without leaving the text box.

Key highlights

• Suggests from the top 100k Danbooru and e621 tags by post count (as of November 2022)
• Completes wildcards, Extra Networks filenames, and custom JSON “chants” for longer prompt presets
• Shows live preview images for LoRA, LyCORIS, embeddings, and hypernetworks inline
• Supports tag translation via community-provided files and optional full-prompt live preview
• Integrates with several third-party extensions including WD 1.4 Tagger and Multidiffusion Upscaler

Caveats

• Default tag lists include NSFW terms; the README carries a responsible-use disclaimer
• LoRA trigger-word insertion can occasionally misfire due to hash collisions
• Some features, like automatic LoRA keyword discovery, rely on an additional extension unless you’re running WebUI v1.5.0 or newer

Verdict

Worth installing if you spend more than five minutes tweaking Stable Diffusion prompts and want the model to actually understand your intent. Skip it if you generate exclusively with natural-language prompts on models that never saw a Booru tag in training.