Prompting for Better Character Results

Booru Basics and Character Prompt

Use comma-separated tags like 1girl, silver hair, blue eyes. Skip long sentences and vague words.

Booru tags are common training labels, so the model reads them more reliably than prose.

Keep tag lists focused. Extra tags that do not matter to the shot usually add noise.

Character tags apply to every image of that character. Put identity traits here: hair, eyes, body type, and permanent accessories.

Do not put pose or framing in the character prompt. Use those in the action prompt.

Filtered mode

Filtered mode accepts recognized booru tags only. A tag that is not in the tag list, including a misspelling, is blocked. Pick a suggestion from autocomplete.

LoRAs are unavailable in Filtered mode. Switch to Unrestricted to use a LoRA or custom text.

Outfit Prompt

Outfit tags are added on top of character tags, not replacing them. Avoid repeating character tags unless you want stronger weighting.

Be specific for consistent clothing. For example: school uniform, blue sailor collar, blue skirt, red ribbon, black thighhighs, midriff.

Once an outfit is created for a character, you cannot edit its outfit prompt.

These two examples are of Kiss with the same outfit prompt: casual, t-shirt

We get very different results because the prompt is not specific enough.

Action Prompt

Action tags should describe pose, framing, mood, or interaction. Because actions are reusable across outfits, avoid clothing tags unless the action needs them. If unsure, start with cowboy shot.

Keep emotion and movement tags in the action prompt. If you place smile or running in character tags, they will appear in most images.

Here is an example of a good action prompt:

cowboy shot, facing viewer
valentine, heart-shaped box, holding gift, incoming gift
averting eyes, embarrassed, blush

Advanced Concepts

More elements in one image usually means less detail per element. Full-body shots often reduce face detail compared with upper-body shots.

Some tags, such as thighhighs, can push composition toward full-body framing.

Square-bracket prompting, such as [thighhighs:0.2] or [tag::0.2], is not supported yet.

Parenthesis weighting can force a weak tag. Example: if an image keeps ignoring closed eyes, try (closed eyes:1.3). A group can share one weight, such as (blue eyes, red hair:1.2).

Use weighting carefully. High values can make results unstable, so keep weights modest and use them only when needed.

Coming soon: Selective removal helps when an action conflicts with default tags. Example: a character has heterochromia, blue eyes, yellow eyes, but the action is closed eyes. Removing eye-color tags for that action prevents those tags from fighting the closed-eye request.

Useful Tags

We will color tags based on their category:

facing viewer, looking at viewer,
lying, on back, m legs,
pussy, clitoris, 1boy, penis,
pillow grab, large insertion, sex, 
:o, blush,

facing viewer, looking at viewer, pov,
reclining, leg up,
large insertion, pussy, clitoris, barefoot, soles, 1boy, penis, sex,
blush, light smile

sitting, leaning back, looking at another,
large insertion, pussy, clitoris, feet, legs apart,
1boy, faceless male, penis, sex, vaginal
blush, @_@, clenched teeth, tearing up, uneven eyes

reclining, head out of frame, from side, looking at another,
hands up, m legs, leg up, bound wrists,
red rope, pussy, clitoris, feet, holding another's leg, 1boy, penis, sex, standing sex, vaginal, motion lines
blush, open mouth, upper teeth only, heavy breathing

leaning back, facing viewer, pov,
(cowgirl position, girl on top:1.1), m legs,
pussy, clitoris, 1boy, penis, testicles, deep penetration, sex
blush, :o, closed eyes, pain

Coming Soon

All images on the website are portrait format at the moment, with no exceptions yet.

Selective tag removal in action prompts is coming soon and will make conflicting cases easier to handle.