Image Prompting & Tagging Guide

Getting exactly the image you pictured is a skill, and it comes down to one thing: how you write your tags. Xotic AI's image models use the Danbooru tagging system, so this guide is about writing good tags — not sentences. Ten minutes here will save you hundreds of wasted generations.

Tags, not sentences#

The models were trained on images labelled with Danbooru tags — short, comma-separated keywords — so that's the language they understand best. Write a list of tags, not prose.

✅ 1girl, solo, long blonde hair, blue eyes, red dress, standing, bedroom, soft lighting
❌ A beautiful woman with long blonde hair wearing a red dress standing in a bedroom

Both might work, but the tag list gives you far more precision and control. A few conventions:

• Separate every tag with a comma. Each tag is a distinct instruction.
• Use Danbooru-style tags. Counts like 1girl, 2girls, solo; descriptors like long hair, blue eyes, thigh highs.
• solo is one of the most useful tags — it keeps the image to a single subject. Add it whenever you want just her.

Structure your prompt#

Order matters. Tags near the front carry more weight, and grouping related tags together helps the model use them. A reliable structure:

1. Quality tags — masterpiece, best quality (see below).
2. Subject & count — 1girl, solo.
3. Body & appearance — age look, body type, large breasts, skin.
4. Hair & face — long blonde hair, blue eyes, smiling.
5. Clothing — red dress, thigh highs (or nudity tags — see the NSFW reference).
6. Pose & action — standing, looking at viewer, arms behind back.
7. Setting — bedroom, on bed, window.
8. Camera & lighting — from above, close-up, soft lighting.

You don't need every section — but this order keeps your most important tags up front where they count.

Quality tags#

Quality tags nudge the model toward cleaner, higher-fidelity output. They matter most in anime mode, where the studio automatically starts your prompt with masterpiece, best quality. Common ones:

masterpiece, best quality, highres, detailed, absurdres

The negative prompt#

The negative prompt lists what you don't want. It's the main tool for cleaning up anatomy glitches and artifacts. A solid, general-purpose negative to start from:

lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits,
cropped, worst quality, low quality, jpeg artifacts, blurry, watermark, text, signature

Add specific things you keep getting and don't want — e.g. glasses, hat — to the negative to push them out.

Emphasis & weighting (advanced)#

This is the power-user technique. You can make any tag stronger or weaker so the model pays more (or less) attention to it.

Parentheses#

Wrapping a tag in parentheses increases its weight by ~1.1×, and nesting multiplies:

(red dress)      → ~1.1× emphasis
((red dress))    → ~1.21× emphasis   (1.1 × 1.1)
(((red dress)))  → ~1.33× emphasis

Explicit numeric weight#

For precise control, put a number after a colon. This is the cleaner, more readable way to do the same thing:

(red dress:1.5)   → 1.5× — strongly emphasized
(red dress:1.2)   → 1.2× — a noticeable boost
(red dress:0.7)   → 0.7× — de-emphasized, model cares less

• Above 1.0 emphasizes; below 1.0 de-emphasizes.
• (tag:1.1) is the same as (tag), and (tag:1.21) is the same as ((tag)) — the numeric form just lets you dial it exactly.

Practical use#

Reach for weighting when a tag is being ignored or overpowering the image:

1girl, solo, (red dress:1.4), long blonde hair, (looking at viewer:1.2), bedroom

Worked examples#

Simple portrait

masterpiece, best quality, 1girl, solo, long black hair, brown eyes,
white blouse, smiling, looking at viewer, upper body, soft lighting

A specific scene

masterpiece, best quality, 1girl, solo, (red cocktail dress:1.3), red lips,
sitting, crossed legs, luxury bar, night, bokeh, cinematic lighting, from side

Dialing in a detail that keeps getting missed

1girl, solo, (freckles:1.3), (green eyes:1.2), curly red hair, denim jacket, outdoors

Common mistakes#

• Writing sentences instead of tags — you lose precision.
• Too many tags — a giant prompt dilutes everything; keep it focused.
• Contradictions — long hair, short hair confuses the model.
• Over-weighting — see the warning above.
• Fighting the character — a character's baked-in appearance (from creation) is already applied; prompt the scene, outfit, and pose, not a different face.

Going explicit#

For adult images, Xotic AI's models understand the full Danbooru NSFW vocabulary. See the NSFW tag reference for a categorized list of explicit tags and how to combine them with the weighting above.

Further reading#

The community guide that inspired parts of this page — Civitai's Prompt Crafting Guide — is a great deeper dive on how tokens and prompts work.