SDXL Anime Character Prompts: Tag Order & Presets

SDXL anime character prompts fail when tag order is random

Most “anime girl, beautiful, masterpiece” rolls look generic because SDXL anime character prompts need a stable tag spine—not a pile of quality spam. On browser workflows such as this site’s Stable Diffusion online generator, stable diffussioncn (stablediffussion.cn) rewards the same habit local anime checkpoints use: put subject count and identity first, then look, clothes, pose, scene, and only then quality tags.

This pack is for SDXL / anime-finetune style text-to-image in the browser. It is not a local WebUI install guide. Paste the templates below, lock a Seed, and change one keyword group at a time.

Booru tag order that keeps identity stable

Anime-trained SDXL checkpoints (and many community finetunes) were captioned in Danbooru-like order. Fight that order and the model “forgets” who the character is.

Recommended spine for anime prompt tags:

1. Subject count — 1girl / 1boy / 1other (or 2girls when intentional)
2. Identity — character name + series/copyright when you want a known look
3. Rating / safety cue — e.g. safe when your UI exposes it
4. Appearance — hair, eyes, age band, body framing (upper body, full body)
5. Clothing — garments before vague style adjectives
6. Pose & expression — looking at viewer, sitting, holding cup
7. Scene & light — indoor/outdoor, time of day, simple backdrop
8. Quality tags last — masterpiece, best quality, absurdres (or the checkpoint’s preferred set)

Rule of thumb: identity before wardrobe; wardrobe before atmosphere; atmosphere before “masterpiece.”

Core keyword sets (mix, don’t spam)

Use one set from each row. More rows ≠ better—conflicting clothes and poses are the usual break.

Layer	Example tags
Count + framing	1girl, solo, upper body
Hair / eyes	long black hair, blue eyes, bangs
Outfit	school uniform, sailor collar, pleated skirt
Pose	looking at viewer, slight smile, standing
Scene	classroom, soft window light, depth of field
Style anchor	anime screencap, cel shading, vivid colors

Natural-language hybrids work on base SDXL, but for anime finetunes, comma-separated tags usually win. Keep English tags even when your UI is localized—most anime weights expect them.

Copy-ready SDXL anime character prompts

School-uniform upper body

1girl, solo, upper body, long black hair, blue eyes, bangs,
school uniform, sailor collar, red ribbon, looking at viewer,
soft smile, classroom, soft window light, anime screencap,
cel shading, masterpiece, best quality, absurdres

Casual outdoor full body

1girl, solo, full body, silver twin tails, green eyes,
hoodie, denim shorts, sneakers, walking pose, cherry blossom path,
spring afternoon, soft sunlight, anime style, detailed background,
masterpiece, best quality, highres

Dynamic three-quarter action

1girl, solo, cowboy shot, short orange hair, amber eyes,
sports jersey, fingerless gloves, dynamic pose, wind in hair,
stadium lights, motion blur background, anime screencap,
sharp lineart, masterpiece, best quality, absurdres

Start vertical (e.g. 832×1216) for portraits; switch to square only after the face and outfit stabilize. More scene packs live in the prompt library.

Anime negative prompts that actually help

A short, targeted negative beats a 40-term dump. Start here:

lowres, bad anatomy, bad hands, missing fingers, extra digits,
extra limbs, fused fingers, poorly drawn face, mutated hands,
blurry, watermark, text, logo, realistic photo, 3d render,
western cartoon, deformed eyes

• Want cleaner faces → keep poorly drawn face, deformed eyes.
• Holding props → keep hand terms; don’t hide both hands in every shot forever—vary composition instead.
• Photoreal bleed on an anime checkpoint → add photorealistic, realistic skin pores.

Deeper negative craft: Stable Diffusion negative prompts guide. Weighting syntax: prompt weighting guide.

Online presets (SDXL anime)

Item	Suggestion	Why
Aspect	832×1216 or nearby vertical	Faces + outfits read cleaner
Steps	28–34	Anime linework needs a few extra steps
CFG	5–7	Higher CFG often fries cel colors
Sampler	One stable default (e.g. DPM++ / Euler)	Don’t swap sampler while debugging tags
Seed	Lock after a good scout	Required for fair tag A/B

If the face collapses after you add clothes, lower CFG 0.5–1 before rewriting the whole prompt. Parameter theory: online generation params guide. Base choice when you also try SD 3.5: SD 3.5 vs SDXL.

Tips that keep characters consistent

1. Never front-load quality tags on anime finetunes—put them at the end.
2. One identity change per run: hair or outfit or backdrop—not all three.
3. Prefer upper body while scouting faces; move to full body only after the look locks.
4. Hide complex hands early (hands in pockets, holding bag behind back) until composition is right.
5. A/B with a fixed Seed so you know which tag group actually moved the image.

Iteration checklist

1. Paste the school-uniform prompt; generate 2–4 images; name the main failure (wrong age band? mushy face? photoreal bleed?).
2. Identity drifts → cut scene fluff; reinforce hair/eyes/outfit before adding more quality tags.
3. Plastic / oversaturated → lower CFG; trim vivid stacks; strengthen negatives for oversaturated.
4. Bad hands → simplify pose first; then tighten hand negatives; use img2img only after the body pose is acceptable (img2img denoising guide).
5. Save “tag spine + negative base” as your anime preset; next session only swap outfit or scene.

Next steps

Run the pack in Generate or adapt cousins from the prompt library. Once tag order feels automatic, you have a reusable SDXL anime character workflow—more durable than collecting one-off “magic” lines.