Babe wake up, new feet checkpoint just dropped!!

Only reason to still be on this site, other then to watch Rod lose his mind.

V6 is here, time to fill the gallery with femboys 🤭

Крутая модель. Но возможности купить нету.

Damn, 14 days of edging, XD

v50运行正常，切换成v60后，参数都不变却无法生成图像，提示是这样的，应该怎么办？：modules.devices.NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enoughprecis

ion to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer

to float32" option in Settings > Stable Diffusion or usingthe --no-half commandline argument to fix this. Use --disab1-nan-cheg

k commandline argument to disable this che

Changelog for v6? :3

The new version 6 turned out interesting. The images seemed to have more contrast, and the shadows were improved, which even forced me to change my usual lore settings because the images were becoming too dark. I spent the whole day experimenting with styles to highlight the advantages of the new model, and I was very pleased with the result. It's nice to see that new versions of my favorite model always have a lot of new and initially unusual features, but the results are worth it.

Happy to early bid, did not know there was a ver 5. i use v4 a lot on my rig. one of my favorite checkpoints by far. looking forward to see what this can do.

The model looks wonderful hope it gets released for free soon

Ohhh the newest gacha waifus are added!

Can I ask why in the desc you insisting no need for upscaling when using your ckpt?

Lets see what you have ....

Was the NSFW version of 4 removed?

Is JANKU_V6 still based on Illustrious_V0.1?

伟大，无需多言

Compared to the previous version, video card memory consumption has been reduced by at least 0.7 GB. Background objects have also improved.

Nice. But can it render faces lit from the front tho?

v6 is definitely an upgrade, and while I think the model is great and love how well it sticks to my prompts, I still feel like every image comes out way too bright. When I compare it to Plant Milk, WAI, or Prefect, JANKU just feels like it’s missing that dynamic lighting those others pull off so well.

Thank you for the model, any way to see character list without discord?

Hi bro i know its to much to ask but can you also make a workflow or share something you used in past

Can you give training data of v6 in terms of character and artists tags as you provided with v5 ? Tht will help a lot in crafting prompt.

I like your model better thn others bcoz it's simple and efficient. Thnkx

Thank you for the model, can you share character list? i have trouble with discord.

One of the best checkpoints i have used, very good!

any chance we could get a list of the new characters added to the training data i cant seem to find it on the discord

Is there a Workflow for comfyui?

Feels like V6 has less anime-like faces, kinda like what happened between LustrousMix V1 and V2. Was this intended?

v6 is definitely an upgrade, and while I think the model is great and I love how well it follows my prompts, I still feel like every image comes out way too bright. When I compare it to Plant Milk, WAI, or Prefect, JANKU feels like it severely lacks dynamic lighting.

Skin tone turn a bit red at last step, Vae are baked in, does anyone know what's the problem??? I tested clip skip 1 and 2, both turn red at last step.

Why do people add videos for image models? It's like showing an airplane when i'm trying to shop for bikes, I hate it

I noticed I'm having difficulties generating images on Comfy with certain styles. As to why, I am not sure.. I've jumped back and forth between a few workflows and the results appear the same with the particular tags. Genning on site, however, the resulting image looks exactly like the art style in question.

Last Update in Tensor please

dont know why i was banned on discord, i didnt even say anything

Trying to do some animations but it's not working with any of my motion modules, i do only have only a few, any recommendations?

Which hi res fix and settings use with this checkpoint?

The effect of artist styles become stronger than previous version, which is good. But it seems be too strong in stylish sometime, make it feels like a little bit unstable when using artist prompt. Such as blush/artist signiture/blurry eyes or other parts, that not set high-priority or even exist in your prompt, and you can not control them easily.

The best balance version is still 4.0 in my mind.

BTW, one of the reason that I usually don't use lazy embeddings is, it may weaken your control over prompt. For instance, black or tun skin prompt turn into white, remove face freckles you wanted and so on. Maybe the embeddings will 'fix' the picture, and some part of your prompt may be considered as need-to-fix.

did anyone else got banned form the discord? i was banned as soon i clicked the "day" challenge, so that was annoying

After playing with it for couple days, I think I like the v5 more than v6.
Since I tend to seek an output that looks like anime screenshot, V5 give better output for my taste.
v6 looks more like 3D-ish anime to me, which not suitable for me.

Good model, Is this the only person who actually update the dataset of the model? If there is someone more who update it please let me know guys.

I was accidentally banned from the Discord channel due to a mistaken operation. Could you please guide me on how to get an invitation to rejoin the channel?

cool

This is my favorite model; however, after V4, many Loras lose quality compared to the current model. It certainly offers incredible quality, but in several cases, it loses that "anime" essence.This is my favorite model; however, after V4, many Loras lose quality compared to the current model. It certainly offers incredible quality, but in several cases, it loses that "anime" essence. Is there no way to enable V4 as well? It worked wonderfully with many LoRas, especially those that weren't as up-to-date, but now even with LoRas from this same account, it's giving undesirable results.

I keep getting slightly translucent clothing in sensitive/nsfw images. Mostly bellybutton and nipples. Anyone else?

Sorry, I keep generating a lot of images with extra fingers. How do you avoid this?

This is one of the few models that knowing of aglaea \(honkai: star rail\) appeared in February 2025.

There are many other models, but don't knowing the new character means that just created a cocktail by merging with other models rather than learning the model.

However, this model proved that creator did effort extra learning.

I deleted every models that don't know aglaea. because making just a cocktail is not only easy but also burying real creators.

thank you very much. and keep go on.

This is good work!

Is there V-pred model ?

Is it just me or are the faces slightly washed out and blurry now? I can't really tell haven't generated in a bit but maybe the new model changed it?

I just want to mention that with version 6, the eyes and fingers are of very poor quality; in version 5, using the same prompts, I get better results.

First off, I want to express my huge appreciation for this model's unparalleled ability to follow artistic style and gesture prompts. After extensive testing, this version stands out as the best at capturing and reproducing specific artist styles—many of which other models simply ignore. That said, there is a new issue I’ve noticed: in certain styles, the eyes (face) can become quite blurry and lose definition, it seems have a little worse stability（such as finger problem）.

Btw, though this version have some obvious issues, I still think it a right step. I had some reservations to previous version because I thought the stylization can be so "overreactive" that it occasionally surpasses the original artist's intensity (sometimes leading to overly sharp lines or slightly unstable compositions). However, after long-term use and comparison, I’ve realized this path is correct. This exceptional prompt adherence is its greatest strength, and this version makes it better.

In fact, despite the eye issues, this v60 has already become my daily use because its prompt-following capability is really better in my test. From my personal experience, the overall stylish ability has also improved significantly. It will be really good model if fix the issues. Keep up the amazing work, this is the right step.

Sometimes the illustrations look yellowish and burnt. How can I fix this? I don't know why, but this problem doesn't occur when I use version V4; it happens in both versions V5 and V6. I've had no choice but to go back to using only version V4.

A lot more vibrant (v6.0) over WAI (v16.0) with a different art style.

Version v6 isn't bad; the style is very good and the result is beautiful. What many complain about is that it tends to generate images with extra or deformed fingers, and even a reddish tone on the faces. In my tests, I noticed that if you don't use LazyNSFW, the problem of the reddish faces disappears. Regarding the fingers, even using LazyHand and a robust prompt, it still generates extra or deformed fingers. Tests with the author's personal resolution of 1024 x 1536 showed the lowest error rate for the fingers, and as for the poses, simply inverting them also works well.

Very cool. Thank you for your work. I really like it