“Anime girl” is one of the easiest subjects to generate and one of the easiest to ruin with the default “AI Instagram face.” Below are 10 style-specific templates plus the 6-layer prompt structure (subject / style / composition / lighting / lens / aspect) that lets you tune them without losing control. The prompts are model-agnostic, but they were re-tested in June 2026 on the current anime engines: Midjourney’s Niji 7 (released January 9, 2026), Black Forest Labs FLUX.2, NovelAI Diffusion V4.5, and the Illustrious / Pony SDXL family on local Stable Diffusion.
TL;DR
- A reliable prompt has 6 layers — subject, style, composition, lighting, lens/material, aspect — each on its own line.
- Pick exactly ONE style anchor (KyoAni, Shinkai, xianxia ink-wash, 90s retro). Mixing style tags produces mud.
- Niji 7 is more literal than Niji 6, so “vibey” one-word prompts behave worse, not better — be specific.
- For an Anlas-free anime workflow, NovelAI V4.5’s Opus tier ($25/mo) generates single images at no token cost; for the cheapest entry, Midjourney Basic is $10/mo (~200 fast images).
- Fix anatomy with an inpaint pass, not with
no extra fingersin the prompt.
Which model should you use?
All ten templates below run on every engine, but the defaults and economics differ. Pricing and versions are current as of June 2026.
| Engine | Current anime model | Entry price | Best for |
|---|---|---|---|
| Midjourney | Niji 7 (--niji 7) | Basic $10/mo, ~200 fast images | Cleanest line art, fastest to a usable result |
| NovelAI | Diffusion V4.5 Full | Opus $25/mo, 10,000 Anlas + free single gens | Booru-tag control, up to 6 characters per scene |
| FLUX.2 | FLUX.2 [pro] / [dev] | API or [dev] open weights (non-commercial) | Photo-to-anime restyling, in-image English text |
| Stable Diffusion | Illustrious XL v2.0 / Pony V6 XL | Free (local GPU) | Full control, LoRAs, no per-image cost |
Notes that matter in practice:
- Niji 7 sharpens eye highlights and pupil structure and follows prompts more literally than Niji 6, so it rewards precise tokens and punishes broad ones. Add
--niji 7on Discord or the web app. - NovelAI V4.5 is its own architecture (not a Stable Diffusion fork), takes prompts up to ~512 T5 tokens, and lets Opus users generate single, non-base, normal-size images with no Anlas cost — the most predictable bill for heavy anime output.
- FLUX.2 [dev] open weights are non-commercial; if you sell the art, use FLUX.2 [pro] via API or a licensed host.
- Stable Diffusion: the 2026 community standard for anime is the Illustrious line (WAI-Illustrious, NoobAI-XL) and Pony Diffusion V6 XL — both use booru-style tags, so the templates below transfer with light token weighting.
What a high-quality prompt should contain
A reliable anime-girl prompt — no matter the model — has these 6 layers. Putting each on its own line gives noticeably better results than mashing them into one sentence:
- Subject: age, hair color, hairstyle, eye color, clothing, accessories — concrete locks the face
- Style: KyoAni / Shinkai / Chinese animation / 90s retro / dark fantasy — pick ONE, never mix
- Composition: upper body / half body / full body / close-up, plus angle (front / three-quarter)
- Lighting: soft window / golden hour / neon rim — single biggest tier-up lever
- Lens / material: 85mm portrait, shallow depth of field OR cel shading, painterly shading
- Aspect:
--ar 2:3for character art,--ar 9:16for phone wallpaper,--ar 16:9for key visuals
10 copy-ready prompt templates
1. Fresh school girl
Best for: Avatar art, light-novel covers
anime portrait of a 17-year-old Japanese high school girl, soft brown twin-tail hair, large hazel eyes, gentle smile, white sailor uniform with red ribbon, modern cel shading, Kyoto Animation style, soft window light, half body, --ar 2:32. Cyberpunk girl
Best for: Game splash art, short-video opener
anime portrait of a 16-year-old girl with platinum white hair shaved on one side, sharp violet eyes, faint scar on cheek, black techwear jacket with glowing cyan circuit lines, modern anime key visual, magenta and cyan neon rim light, --ar 16:93. Magical girl
Best for: Fan art, kawaii covers
anime full-body shot of a 14-year-old magical girl, twin-tail pink hair, sparkling amethyst eyes, ruffled white-and-pink dress with star accents, magical wand raised, classic shoujo anime, pastel rainbow palette, sparkles, --ar 2:34. Shinkai high schooler
Best for: Key-visual posters, TikTok cold open
anime portrait of a 17-year-old girl in summer uniform standing on a rooftop, soft wind in her hair, looking up at the sky, Makoto Shinkai style, hyper-saturated cinematic key visual, golden hour clouds, lens flare, ultra-detailed background, --ar 16:95. Xianxia swordswoman
Best for: Novel covers, Chinese-style fan art
anime portrait of a 20-year-old swordswoman in flowing white robes with red trim, long black hair tied half-up, calm focused expression, ornate silver sword at her side, modern Chinese animation, ink-wash background, --ar 2:36. 90s retro girl
Best for: Retro key art, album covers
anime portrait of a 19-year-old woman with short bob cut and red lipstick, oversized denim jacket, yellow scarf, 1990s anime aesthetic, slight film grain, muted warm color palette, retro cel shading, --ar 4:57. Chibi sticker girl
Best for: Stickers, emoji packs, app icons
chibi anime girl, big head small body proportion, cute round eyes, hoodie with cat ears, plain pastel background, kawaii sticker design, clean vector look, --ar 1:18. Dark fantasy mage
Best for: TRPG character sheet, fantasy covers
anime portrait of a 22-year-old female mage, long silver-grey hair, intense violet eyes, faint scar across left brow, black hooded robe with gold runic embroidery, dark fantasy painterly shading, --ar 2:39. Summer festival girl
Best for: Summer posters, event art
anime portrait of a 16-year-old girl in a yukata holding a red paper lantern at a summer festival, warm orange evening light, fireflies in background, soft painterly anime style, --ar 4:510. Noir detective
Best for: Fan art, mystery covers
anime portrait of a 25-year-old female detective in a beige trench coat, sharp focused expression, dim noir bar lighting, cigarette smoke haze, rain on window, painterly cel shading, --ar 4:5Common mistakes
- Writing only
cute anime girl, beautiful— the model has no anchor, every roll drifts. This hurts more on Niji 7 than Niji 6 because Niji 7 reads prompts more literally. - Mixing
Shinkai style+90s retro+cyberpunkin one prompt — styles collide into mud - Stuffing every layer into one sentence — the model under-weights the tail
- Forgetting aspect — defaults to 1:1 and crops legs / hair off the character
- Trying to fix anatomy in the prompt (
no extra fingers) — far less reliable than an inpaint pass at ~1MP
How to push results further
- Reuse a shared “style header” (the style/lighting/lens lines) across a series; only swap the subject line for visual consistency
- Midjourney Niji 7: pair with
--sref [image-url]and--sw 100to lock the look beyond pure text; Niji 7’s sref handling was upgraded over Niji 6 - Weight critical tokens so they survive prompt compression — on Illustrious/Pony SDXL use
(red ribbon:1.3); on Midjourney repeat the token - Lock a seed for iteration — Midjourney
--seed 12345; NovelAI and every SD UI expose a seed field too - Generate at the model’s native resolution, then upscale with ESRGAN or Topaz instead of asking the model for “8K”
- On NovelAI Opus, keep generations single (no base image, normal size, steps ≥ 2) to stay in the Anlas-free lane
FAQ
Q: Should I prompt in English even on a Chinese SD UI?
A: Yes. Diffusion training captions are overwhelmingly English; English prompts are more stable everywhere.
Q: Why don’t my faces match the templates?
A: Usually extreme aspect (21:9 warps faces), an outdated model version, or weak negatives. As of June 2026, switching to --niji 7 on Midjourney, NovelAI V4.5 Full, or an Illustrious XL v2.0 checkpoint fixes most of these. Niji 7 specifically improved eye highlights and pupil structure over Niji 6.
Q: Can I blend two anime styles?
A: Text-mixing rarely works. Use dual --sref references on Niji 7, NovelAI’s style-mixing (improved in V4.5), or generate single-style then blend with IP-Adapter on Stable Diffusion.
Q: Which model is cheapest for high-volume anime output?
A: NovelAI Opus ($25/mo as of June 2026) — single, non-base, normal-size generations cost zero Anlas, so heavy output is effectively unmetered. Midjourney Basic ($10/mo, ~200 fast images) is the cheapest entry but meters every image. Local Stable Diffusion (Illustrious/Pony) has no per-image cost if you own the GPU.
Q: How do I keep junk out of the background?
A: simple background / plain pastel background + drop vague atmosphere words (“mysterious atmosphere”). Be specific, not evocative.
Q: My prompts trip the NSFW filter — fix?
A: Swap swimsuit / bikini / stockings for neutral terms (summer outfit, athletic wear); add safe for work, portrait illustration.
Related articles
- Anime Boy Image Prompt Examples
- Realistic Beautiful Woman Portrait Prompts
- Game Character Design Prompt Examples
- Fixing distorted AI faces
- Back to Prompt Library
External references: Midjourney plan comparison, NovelAI image models, Black Forest Labs FLUX models.