Illustrious, NoobAI and Pony prompting: tags, quality strings, artist tags and LoRAs
Subiect: Prompting
De Captain
Publicat 2026-05-30
The tag-based prompting that anime SDXL models expect: Danbooru tag order, quality and year tags for Illustrious and NoobAI, score tags and source tags for Pony, artist tags and how strongly to weight them, negatives that work, and which LoRAs fit which base.
Această pagină nu a fost încă tradusă, așa că este afișată în engleză.
Prezentare generală
Illustrious XL, NoobAI-XL and Pony Diffusion V6 are
SDXL
fine-tunes trained on booru-tagged images. They do not want sentences; they want comma-separated tags in a particular order with quality strings in front, and each base has its own strings. Get the strings right and these models produce the cleanest anime and stylised art of any open family; get them wrong and results look like generic SDXL.
The three bases are not interchangeable for
LoRAs
. Pony LoRAs expect Pony's score tags; Illustrious and NoobAI share Danbooru tags and many LoRAs cross between them with small quality loss. On this site the tag row filters SDXL models by illustrious and ponytype so you can keep each ecosystem straight.
Anima
, a newer non-SDXL base, uses the same tag vocabulary and has its own family page and artist-tag guide.
Everything here runs on 8 GB locally. It also runs hosted:
PirateDiffusion
renders these checkpoints with /render and the trigger shown in the Run box, with recipes like #sdxlreal that add the quality strings for you;
BitVector
has them in the web app.
Referință
Nume | Tip | Ce este |
|---|---|---|
Tag order | prompt | Quality tags, then count (1girl, 2boys), character, series, artist, then appearance, clothing, pose, setting, composition, style. Weight falls off toward the end. |
Illustrious quality string | prompt | masterpiece, best quality, very aesthetic, absurdres (v1 and v2); add newest or year tags (year 2024) for modern style. |
NoobAI quality string | prompt | masterpiece, best quality, newest, absurdres, highres (EPS); v-pred versions need the v-pred checkpoint settings. |
Pony score tags | prompt | score_9, score_8_up, score_7_up, score_6_up plus source_anime / source_cartoon / source_furry / source_pony and rating_safe / rating_questionable. |
Negative string | prompt | Illustrious: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digits, worst quality, low quality, jpeg artifacts, signature, watermark, username, blurry. Pony: score_6, score_5, score_4, plus the same. |
Artist tags | prompt | Danbooru artist names (by \(artist\) when ambiguous). 0.6 to 1.0 weight; two or three blend styles. |
Character tags | prompt | Danbooru character name \(series\) renders known characters without a LoRA on Illustrious and NoobAI. |
Settings | setting | 1024x1024 or 832x1216; Euler a or DPM++ 2M Karras; 25 to 30 steps; CFG 5 to 7 (Pony 6 to 7, NoobAI v-pred 4 to 5); CLIP skip 2. |
Escaping | syntax | Parentheses in tags must be escaped: hatsune_miku_\(vocaloid\). Underscores and spaces both work on modern bases. |
AD

Fără instalare - rulează modele AI în cloud
BitVector Prism este cel mai simplu mod de a începe: alege un model, tastează un prompt și generează într-o aplicație web curată, fără nimic de configurat. BitVector este disponibil și pe Discord și pe web (SpyGlass).
Pas cu pas
- Pick a base on the SDXL family page using the illustrious or ponytype tag filter, download it, and set CLIP skip 2.
- Start the prompt with the base's quality string, then the count tag, then the subject.
- Add appearance and clothing as tags, then pose and expression, then background and lighting, then composition tags (cowboy shot, from above, dutch angle).
- Add one artist tag at 0.8 for style, or a style LoRA from the same base at 0.7; not both at first.
- Use the base's negative string. Generate four images at 1024x1024, Euler a, 28 steps, CFG 6.
- Adjust: anatomy issues, add bad hands negatives and use Adetailer; flat style, add a second artist tag; too much style, lower artist weights.
- Character LoRAs: trigger word plus the character tag together, LoRA at 0.8; the model page lists both.
- Hosted: /render <checkpoint-trigger> masterpiece, best quality, 1girl, ... [[lowres, bad anatomy]] on PirateDiffusion, or the web app on BitVector.
Exemple
Illustrious
masterpiece, best quality, very aesthetic, absurdres, 1girl, solo, silver hair, long hair, red scarf, winter coat, snow, shrine, torii, dusk, lantern light, looking at viewer, cowboy shot, (artist:0.8)
Negative: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digits, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, username, blurry
Pony
score_9, score_8_up, score_7_up, source_anime, rating_safe, 1girl, solo, silver hair, red scarf, snowy shrine, dusk, lanterns, looking at viewer, cowboy shot
Negative: score_6, score_5, score_4, bad anatomy, bad hands, text, watermark
PirateDiffusion
/render <illustrious-xl> masterpiece, best quality, very aesthetic, absurdres, 1girl, silver hair, red scarf, snowy shrine, dusk [[lowres, bad anatomy, bad hands, watermark]] /size:1024x1024 /steps:28 /guidance:6 /adetailer
Sfaturi
- The Anima artist-tag guide on this site lists tags that also work on Illustrious and NoobAI, with example strengths.
- Year tags (year 2023, newest) on Illustrious and NoobAI shift toward modern rendering; oldest goes retro.
- rating tags (rating_safe) matter on Pony and help on NoobAI for keeping output clean.
- For realism on these bases use the semi-realistic merges tagged on the family page rather than fighting an anime base.
- Hands: 2 to 3 negatives plus Adetailer hand model fixes most; CFG above 7 makes them worse.
Depanare
Generic, washed-out look
De ce se întâmplă
Quality string missing or from the wrong base.
Cum se repară
Use the base's own string first in the prompt; CLIP skip 2.
Pony output is noisy or grey
De ce se întâmplă
Missing score tags or CFG too low.
Cum se repară
score_9, score_8_up, score_7_up first; CFG 6 to 7.
NoobAI v-pred produces burnt colours
De ce se întâmplă
v-pred checkpoint used with EPS settings.
Cum se repară
Enable v-prediction (ModelSamplingDiscrete v_prediction with zsnr in
ComfyUI
; the v-pred option in Forge) and CFG 4 to 5, or use the EPS version.
Artist tag does nothing on Pony
De ce se întâmplă
By design.
Cum se repară
Style LoRA trained for Pony.
Character wrong or merged with another
De ce se întâmplă
Ambiguous tag or two characters without separation.
Cum se repară
Add the series in escaped parentheses; for two characters use regional prompting or a two-character LoRA.
LoRA degrades quality
De ce se întâmplă
LoRA from a different base.
Cum se repară
Match base (illustrious / ponytype tag on the model page), weight 0.7.
AD

Fără instalare - rulează modele AI în cloud
PirateDiffusion este doar pe Telegram și construit pentru profesioniști: mii de modele, LoRA și fluxuri de lucru controlate prin comenzi de chat, cu generare nelimitată la un plan cu preț fix.
Întrebări
Illustrious or NoobAI or Pony?
Illustrious for the broadest current LoRA library and clean results, NoobAI for the most characters and artists known natively, Pony for stylised and western-cartoon looks and its own LoRA world.
Do these work in natural language?
A little, but tags are what they were trained on. Mixed prompts (tags plus a short sentence) are fine.
Where do I find tag names?
Danbooru's tag wiki; the Anima artist-tag guide on this site; and the example prompts under every model's renders here.
Can I run them on 6 GB?
Yes with --lowvram or Forge; or hosted on PirateDiffusion and BitVector.
Are Anima LoRAs compatible?
No; Anima is a different architecture with the same tag language. It has its own family page.
Linkuri și surse
- Illustrious XL on Hugging Face
- NoobAI-XL
- Pony Diffusion V6 XL
- SDXL family page
- Anima family page
- SDXL on PirateDiffusion (recipes and /render)
- BitVector
Modele din acest ghid
Scris de
Captain
Ghiduri conexe
Prompting
Anima artist tags: the complete @artist and tag guide
How Anima reads Danbooru-style tags, how to call an artist style with an @artist tag, how to mix two or three artists, and a ranked table of the most used Anima artist tags with links to the Anima Style Explorer and AnimaDex.
De Captain
2026-10-03
Modele
What is a LoRA and how to use one in ComfyUI, Forge, Automatic1111 and on the cloud
LoRAs explained for people who want results, not maths: what a LoRA file changes, why it must match the base model's family, trigger words, weights, stacking several LoRAs, the folder and syntax for each UI, and how to try a LoRA on the cloud before downloading anything.
De Captain
2026-05-20
Modele
Flux vs SDXL vs SD 1.5 vs Qwen vs Z-Image: which model family to use in 2026
A practical comparison of the open image model families: prompt following, realism, anime and illustration, text rendering, speed, VRAM, LoRA ecosystem and licence, with a recommendation for each kind of work and hardware.
De Captain
2026-05-22
Erori și soluții
Automatic1111 (Stable Diffusion WebUI) errors and fixes
Fixes for the Stable Diffusion WebUI errors that stop most installs: "Torch is not able to use GPU", NansException, CUDA out of memory, "Couldn't install torch", xformers, "Stable diffusion model failed to load", broken extensions after an update, and the SDXL-specific problems.
De Captain
2026-05-14
Mai multe ghiduri
Modele
FLUX.1 Dev: how to prompt it, best settings and creative ideas
A practical guide to FLUX.1 Dev by Black Forest Labs: natural-language prompting, guidance and step settings, LoRA stacking, text rendering and prompt ideas that play to its strengths.
De Captain
2026-09-30
Modele
Stable Diffusion XL 1.0: prompts, settings and what it still does best
How to get the most out of the official SDXL 1.0 base model: native resolutions, CFG and sampler settings, the refiner, negative prompts and ideas for styles where SDXL still shines.
De Captain
2026-10-01
Modele
Stable Diffusion 1.5: the classic model, prompted properly
Stable Diffusion 1.5 is still worth knowing: the right resolution, CFG and sampler, how to use its huge library of LoRAs and embeddings, and creative prompt ideas suited to a 512-pixel model.
De Captain
2026-10-02
Modele
FLUX.2 Dev: prompting the newest Black Forest Labs model
FLUX.2 Dev brings larger prompts, better text, stronger realism and multi-reference editing. This guide covers the turbo workflow, guidance and resolution settings, prompt structure and ideas that use its new strengths.
De Captain
2026-10-03
Modele
Qwen-Image: long prompts, perfect text and bilingual posters
Qwen-Image is the model to reach for when the words in the picture matter. Learn how to write its long descriptive prompts, render English and Chinese text accurately, set CFG and steps, and explore layout-heavy creative ideas.
De Captain
2026-09-29
Modele
SDXL-Lightning: four-step generation on any SDXL checkpoint
ByteDance's SDXL-Lightning LoRA turns a 30-step SDXL render into a 4-step one. Learn the step counts, the CFG you must use, the sampler, and how to combine it with your favorite checkpoints and style LoRAs.
De Captain
2026-09-30

Model
Trends
.ai
© ModelTrends.ai
|
Realizat în Japonia
|
© 2026