Booru Smart Prompt Generator

Generate semantically coherent Danbooru-style tag prompts. Pick a generation mode and optional content rating, then tune sampling controls.

Mode

Empirical = follow dataset frequencies (alpha=0). Diverse = boost rare tags (alpha=1). Custom alpha lets you set the slider below.

Alpha (custom)

0 = empirical distribution, 1 = maximum diversity. Only used when Mode is 'Custom alpha'.

0 1

Content rating

g = general, s = sensitive, q = questionable, e = explicit. Conditions the model on the rating token.

Number of prompts

How many independent prompts to generate.

Tags per prompt

Target number of tags in each prompt (excluding anchor tags).

1 128

Anchor tags

Comma-separated tags that must appear in every prompt. Unknown tags are ignored.

Blacklist tags

Comma-separated tags the model must not generate. Unknown tags are ignored.

Min model probability

Tags whose raw model probability is below this threshold are dropped. If no tag passes, a top-k fallback is used.

0 0.1

Temperature

Lower = more deterministic / focused. Higher = more random.

0.1 2

Top-k

Keep only the k most likely tags at each step. 0 disables top-k.

0 200

Top-p (nucleus)

Keep the smallest set of tags whose cumulative probability exceeds this value. 1.0 disables nucleus sampling.

0 1

Distribution bias weight

How strongly the empirical/diverse bias is applied on top of the model's scores.

0 2

Seed (-1 for random)

Set a non-negative seed for reproducible generation.

Generated prompts