Waifu Diffusion: Danbooru-tag prompting for a classic anime model

Q: Melted hands and extra fingers

An SD 1.x limitation made worse by low resolution. Add bad hands and missing fingers to the negative, generate at 512x768, and inpaint hands at a higher resolution.

Q: Grey, low-contrast colors

The default VAE in some setups. Load the standard SD 1.5 VAE or the anime kl-f8 VAE and raise CFG to 9-10.

Onderwerp: Modellen

Door Captain

Gepubliceerd op 2026-10-01

Waifu Diffusion is the anime base model that taught a generation how to prompt with Danbooru tags. This guide covers tag order, quality tags, negatives, the right resolution and ideas for clean anime illustrations.

Deze pagina is nog niet vertaald en wordt daarom in het Engels getoond.

Overzicht

Waifu

Diffusion by hakurei is a long-running fine-tune of Stable Diffusion 1.x on tagged anime illustrations. It expects Danbooru-style tags rather than sentences and produces crisp, consistent anime line art and coloring.

It is well-rounded rather than specialized: good at portraits, scenery and simple scenes, less sharp than the newer

Illustrious

family, but light, fast and compatible with the whole

SD 1.5

LoRA

library.

Naslag

Naam	Type	Wat het is
Prompt	tags	Danbooru tags separated by commas: count, subject, appearance, clothing, pose, setting, style.
Negative prompt	tags	lowres, bad anatomy, bad hands, text, error, missing fingers, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, blurry
CFG	7-11	Higher than realistic models; 9 is a good middle.
Steps	25-30	Euler a or DPM++ 2M Karras.
Resolution	512-768	512x768 for portraits; use hires fix for larger.

Geen installatie nodig – draai AI-modellen in de cloud

BitVector Prism is de makkelijkste manier om te beginnen: kies een model, typ een prompt en genereer in een overzichtelijke web-app, zonder iets in te stellen. BitVector is ook beschikbaar op Discord en op het web (SpyGlass).

BitVector proberen

Stap voor stap

Start with masterpiece, best quality, then the subject tags (1girl, solo), then appearance (long black hair, green eyes).
Add clothing and pose tags, then the setting (classroom, rainy street), then a lighting or style tag.
Paste the standard
negative prompt
and leave it alone.
Generate at 512x768,
CFG
9, 28 steps with Euler a.
Upscale with hires fix at 0.5 denoise and an anime
upscaler
for clean lines.

Voorbeelden

Portrait

masterpiece, best quality, 1woman, solo, short silver hair, blue eyes, knit cardigan, looking at viewer, cherry blossoms, wind, soft lighting, detailed background

Scenery

masterpiece, best quality, no humans, scenery, mountain village, rice terraces, morning mist, wooden houses, wide shot, detailed

Creative idea: cozy still life

masterpiece, best quality, no humans, still life, kotatsu, mandarin oranges, sleeping cat, winter window, warm indoor light, soft shading

Tips

Use underscores or spaces consistently; Waifu Diffusion was trained with spaces replacing underscores.
Tag order matters: the first few tags dominate the composition.
Scenery prompts (no humans, scenery, sunset, torii) produce lovely background art.
Combine with an SD 1.5 style LoRA at 0.6 to modernize the look without changing the tag workflow.

Probleemoplossing

Melted hands and extra fingers

Waarom het gebeurt

An SD 1.x limitation made worse by low resolution.

Zo los je het op

Add bad hands and missing fingers to the negative, generate at 512x768, and inpaint hands at a higher resolution.

Grey, low-contrast colors

Waarom het gebeurt

The default VAE in some setups.

Zo los je het op

Load the standard SD 1.5 VAE or the anime kl-f8 VAE and raise CFG to 9-10.

Geen installatie nodig – draai AI-modellen in de cloud

PirateDiffusion is alleen op Telegram en gemaakt voor pro's: duizenden modellen, LoRA's en workflows via chatcommando's, met onbeperkt genereren voor een vaste prijs.