Updated:

After long-term use, I personally think that the current 5.0 version is a bit too focused on the stylized aspect. Style prompts sometimes have a significant impact on the result, and even affect aspects not only the style.

For instance, some artist prompts can lead to a higher likelihood of hairpin/ clothing preference for a certain color/ more intense color prompt pollution/ prompts of artists who are characterized by small eyes may cause overly small pupils/ and so on. Tested with the same prompts in 4.0.

In my personal opinion, 4.0 performs better in maintaining a balance between stylization and stability.

Previous:

5.0 could follow the artist's prompts better than 4.0, but sometimes maybe over-reactive. It still remains the best anime model and has reached a new level. The model itself has very high availability even without the use of embeddings.

Personally, I prefer 4.0 because I like to restore the artist's style with a slight 2.5D feel, but a stronger influence of artist prompt words is undoubtedly an advantage. Though sometimes the stylish may be too strong, it means the artist's prompt words have an overly strong impact on the picture, but the picture‘s style still has a gap from original artists.

Btw, the artist tags will be stronger if you use by+artist name, though you can use artist prompts directly. In my own experience, the effect of only artist_name in 5.0 is close to by+artist_name in 4.0.

There's some other small issues, such as, some previously invalid artist tags are still invalid in this version ; sometimes it still generate blurry or black pussy. Then I feel 5.0's picture stability is also little worse than 4.0, it seems easier to generate watermark and broken part (maybe cause by its stronger stylish mentioned at top).

After all, I like it very much and spent buzz right away. It will be nice to increase the stability part of the model in the future, the strong stylish may break the picture structure and the stability of characters' appearance. Looking forward to your future great works.

Did you complete with ADetailer or is the model already accurate enough?

Wow, amazing bro

Amazing

I really liked this model. :D

Insane 🔥

Unfortunately, there's a problem with this model: it doesn't generate white fishnet thighhighs (or pantyhose) it only makes black fishnet. The other models merged with this one are probably facing the same issue.

I am lucky to be born and live with a person like you, the best checkpoint. I look forward to new versions and new loras from you

The catbox file for artist list isn't working?

Images come out really bright, overexposed and smoothed out like the ones posted in the gallery. With the same loras and negative prompts, a checkpoint like ntrmix doesn't have this problem. How to stop this?

This looks amazing can't wait for this looks like bestest anime checkpoint RIP Im brokie and cant buzz it :D

Best one so far, very balanced.

Works amazing!

Amazing Checkpoint!!

one of the cleanest and highest-quality feet checkpoint ever

cool

OutOfMemoryError: Allocation on device。我的sd一直是最后无法成图的提示

これは素晴らしいです

XLモデルの中でも複数手の指示に対する応答性は最高レベルに近いです

アーティストタグがなくてもキャラクター名を指定すれば公式版に近い形で再現されてる

これは良いです

（日本語で失礼しました）

Yet another amazing release!
Such a great model.

The knowledge of Rouwei 0.8 has been added to include much more concepts. Very good

great model, but for some reason when I added quality tags like masterpiece, best quality, absurd everything turned green, so I don't care but still somehow strange

Where can I find a list of characters? The Discord link isn't working.

There is no point in putting cfg more than 4, and I got the best results before cfg 3. and on 2, the most beautiful and realistic and warm lighting.

Thanks for the improved v5

any prompts structure change? for quality tags, should i add them in the start or end of prompt? also, should i add BREAK to separate them (quality tags) and main prompt?

excellent model!!

I noticed that the image quality is best in UNET bf16, even in fp32 there are more artifacts.

I advise you to double check and add it to the recommendations.

nice

I shared an image on Discord from the creator and they banned me. It wasn't allowed. I'm really sorry.

A good model became the Excellent one

Discord links are inactive.

Decent model. One flaw I've found is that it's pretty much impossible to get dark images without going deep into LORAs.

Umm, I still think the Janku 2.1 style was the best. Any prompt tips on how to get the same effect in v5?

Excellent model that meets my needs~

What is the recommended CFG in this latest version?

just wanted to say hi

Can you please add Tilty Claret from Tenten Kakumei?

V.5 prompting is way better than V4, but still struggle with it on some occasions, i have to add a lot of negative prompts to remove unnecessary details, the style and quality is really great thought.

Hello, I'm a newcomer and don't know much about AI. I'd like to ask, is your model classified as IL or Noob? I noticed that you mentioned NoobAI-XL 1.0 License in the comments, but the title and category column are written as IL, which makes me a little confused about which category it is. If possible, please let me know

If you are trying to get a dark image, try using (((night))) or a similar tag with triple weight. Has worked for me to turn off the lights in most images.

How to generate video? any workflow? plz~~~

Great model, generates amazing details and is really good for feet images, not that I like feet images that much anyways 😖❤️

I’ve been a fan of 'Janku v4.0.'

Although I haven’t had the chance to experiment much with it yet, I feel that the new 'v5.0' version has improved significantly in quality compared to the previous one. The results are amazing.

Initially, I started using this because the output was brighter compared to other checkpoints, but I’ll definitely continue to use this going forward. Thank you as always.

Loving the V5. The only issue I may have is that it keeps generating "comic page" style images, with panels and bubbles showing details of elements... It is an interesting quirk, but sometimes I just want a single image.

Found the reliable solution for shiny bright wet skin that is always persistent in this model. You can either go with some (strong) artist's style or download shiny wet skin lora(https://civitai.com/models/1568999/shiny-wet-skin) for IL and bounce it from -0.5 to -1 depending on how much you like and it would absolutely destroy the "shiny look" while mostly preserving details.

这个模型真的太神了！基本涵盖了所有！包罗万象！堪称2D范围内的永乐大典，内涵的角色数量之多就先不说了，光是这一大堆的绘师就足够研究大半个月了，可以结合出星辰般的组合，神！

"In this version, if you are not using character tags, I highly recommend using lazyloli in negative to make characters more mature."

Thanks, but no thanks! ( ͡° ͜ʖ ͡°)

This is a great model, but for some reason, the characters it generates tend to be on the younger side (it's easy to generate young boys and girls, even when I add lazyloli to the negative prompt, the age still skews young. It seems the training data has a lot of children W_w). I know I can use an age LoRA to control it if I want to generate adults.

This is my favorite checkpoint!

666

Love your work!

I have a question though: in your training files, there is a leafa_(terraria)_(sao), but I can't never get the Leafa Terraria looks at all, it always give me Leafa + the game Terraria related, is there a way to correctly output Leafa with Terraria character look?

I may be stupid, but how do I do characters from the Rouwei list? When I prompt them, the output is not the chars in the list?

Could the master train the fingers and toes better in the next version? Sometimes there are extra fingers or toes, or the hands or feet are very stiff.

Excellent. Thanks for sharing and making it free to use.

Very nice 👍

Incredible and versatile, thank you for this!

Quick question, when I use a higher res, like 1536x1024, I can't seem to keep the number of characters consistent, no nmatter what keywords I use, there's usually always more. Any suggestions?

great

Great model, great tutorial! Reading it definitely hels alot to understand how to use the model + the lazy tools!

生成的图片角色有年龄偏小的倾向，别的都不错

Great model! Love it

this + lazy embeddings + trendcraft all goes HARD. my new favorite model

This is genuinely the best and most cooperative model I've ever used for spicy purposes, and I've tried quite a few of those. Coupled with Smooth Detailer Booster (which adds a little extra "punch" and improves output stability) and a by jtveemo, by asanagi, by modeseven artist combo, the results blew me away.

crazy

I never liked anime... until I downloaded your model. Bravo, amazing!

BEST ANIME MODEL imo

very good

商用利用は可能ですか？

Best model if you are into anime, feet, or both

Thank you for this model 👌😍

yellow thing likeys

Does any one have any advice for prompting with 2 or more characters? Things fall apart when I try to set a scene T~T

Easily my favorite model. Been strictly using it ever since.
No matter how absurd the prompt and weights I throw in, it would always deliver.

What version of Illustrious does JANKU v5 use?

Who can tell me which one the workflow is?

I like this one but... it really struggle to make male character

publib hair is not used.....

and im sad...

Can you create Embedding for neon glowing eyes? Like this

https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1cc3369e-9b40-4d3a-898b-7c7b8343adc2/original=true,quality=90/78653472.jpeg

this model is really good !

Since the buzz policy change, I would like to download your model then try it with Comfyui locally. (I still need to learn about Comfyui anyways.)

Is it OK to use it locally with my 8gb vram RTX2070 and 16gb ram PC or I have to upgrade my PC?

Tried Yodayo these days but the results without your model were all disappointing. Seems I have to learn to use it locally.

I met your model just a couple of days ago... and i can say its already my top 1 favorite. Really amazing and im having a lot of fun creating.

I would like to ask for suggetions or tips in a specific type of composition, im trying not to use specific loras. But to the point, i would like to try to nail a meat chamber, flesh walls, the type of environment you would expect to see in a nest of tentacles. Tentacle pit doesnt work enough because it tries to create a bed or a "hole" of tentacles, is there any keywords or better wording for what im trying to achieve?. I Kinda lack of a super wide english vocabulary and my mind is runing short in concepts xD.

The best model ever!!! I can't stop generating images, the results are soo beautiful, thank you so much ❤️❤️❤️

The Discord link doesnt work..could you share antoher link to the supported character list?

Nice ... but, just another Japanese style model. We kind of need styles from the rest of the world :)

V4 or V5?

Excuse me for asking, but what should we do with the training data file?

大佬们，多眼皮的问题有什么解决方案啊，这个是模型通病吗

This model is one of my favorites of all times so far ♥ i throw any insane prompt at it and it just delivers! Hands, faces, body composition, and popular anime+video games characters! (i use it a lot for Wuthering Waves art) thank you so much for your amazing work and efforts ♥ (i just hate stacking and using loras in general and you solved this) can't wait for your next version! PS for readers: Make sure you use similar settings to the ones recommended by the author (using the recommended resolutions is CRITICAL! for getting good images) 1024 x 1024 is also a good default for 99% of SDXL/ill checkpoints.

what's the recommended upscale from 1024X1536?

V6 just 😍

Instant purchase. ;)

Anyone else having issues with the discord link? I keep receiving "Unable to Accept Invite". Anyone else getting that error?

i just bought. i'm so cool (≖⩊≖)