All models are exclusive to Civitai ! Anyone who publishes my models without my consent will be reported!
Hi all!
SDXL_Niji_Seven is available!
I have incredible results with my LoRa model SDXL_Niji_V6_DLC_LoRa_ v2 at strength 0.5!
https://civarchive.com/models/541460?modelVersionId=601994
I haven't tested v4 yet
What's new?
-This version is very different because it is based on NijiV6 To be more precise, around 1000 images from NijiV5 and 1600 from v6 (compared to a total of 1300 images in v6).
-I chose not to use the "expressive" style of NijiV5 this time. But maybe I will add it in a 7.5 version in the future.
-There is no longer a trigger word
-Version 6 is very different so I still recommend it.
Some tips for use:
-Don't hesitate to play with things like: realistic, hyper realistic, anime studio, digital artwork, illustration...
-Clip skip -2 or -3 is recommended (-3 better overall)
-A minimum of 26 steps
-cfg scale: between 3.5 and 6
-Sampling method: DPMPP_SDE Karras (recommended, best quality), Euler_Ancestral simple (2nd best), DPMPP_2M_SDE Karras or ddim simple (average quality for both).
Have fun ! 😊
Description
FAQ
Comments (32)
Merci beaucoup pour NijiV7 🙏
I thought that it was fun that you made the original announcement in French 😁👍
Yes ! It's easier for me to write it in French first. And I posted the wrong one at first 😂
@Stan_Katayama It was a cool mistake👍😂😎
So it's more like a v6.5?
No ! The dataset is 100% new. And no “expressive” style image which completely changes the visual.
@Stan_Katayama ookay thanks You , i Will give it a try, still Will wait for The Expressive version 👌🏻
Totally worth it !! Thank you for your efforts
will share best creations
Thank you so much ! I'm really dreading the critical reception of this model. The approach is really different! And I already know it will divide 😅
Oops, I accidentally clicked on the new version... Bha will have to test it, too bad <;o)~
Do not force yourself 😅
I have incredible results with my LoRa model SDXL_Niji_V6_DLC_LoRa_ v2 at strength 50!
is that 50 a typo? if not, what is it refering to, cuz it cant be lora strenght 50
Thanks for pointing that out to me. I meant 0.5.
I'll fix that. ☺️
Hi! Good checkpoint. Can I use it for making furry/kemono stuff like Grimoire of Zero furry characters? or has anyone done before?
Hi ! I didn't train it to do that. But if it works, I'm for everyone's freedom to do what they want! 😉
Looking forward to use 7 . Thankyou !! ^_^
thank you for your support 🙏
The model is a little different from the previous ones. It may take a little time to adapt.
@Stan_Katayama Thankyou for making checkpoints like these ♥
@Stan_Katayama Retro is way better on this one , better than katayama. I think this is my new fav model haha
@Image_Alchemy Yes. I needed to do something new. Have fun!
Very good model! I want to train an XL version of the SD model, but the result of the drawing is always easy to yellow, before training SD1.5 version will not be like this, may I ask you have encountered this problem? What causes it? I hope you can give me some experience and help. Thank you
Thank you very much!
Unfortunately I have never encountered this problem and cannot help you on this point. Maybe you should go to the discord server of the software you use? The communities are often nice and helpful.
Sorry
@Stan_Katayama Thanks for your reply! I saw an article on huggingface that basically explains why the SDXL color range is biased towards yellow, https://huggingface.co/blog/TimothyAlexisVass/explaining-the-sdxl-latent-space. If you're interested, you can read this article. Of course, if you are inspired after reading it and it could help me solve this problem, that would be even better!
@qingyuliu1993 Thanks for sharing the article! I hope your problem will be solved now! Have a nice day 😊
Finally I have found my favorite model 💯💗
Many thanks ! 🙏
I have another model that you should like!
It's a merge of SDXL_Niji_7 and 2 other models that I never published. It's called Katayama_Mix_XL_v2 and I personally find it even better
@Stan_Katayama thanks for sharing, I 'll try it!
I just love this model. The only improvement I would like to see is better photorealistic pictures, but if it does not happen, I'm ok with it as is.
For photo style image, you can try https://civitai.com/models/556024/katayamamixxl which is a blend of 50% NijiV7 + Some other good stuff. It is better at photo style images compared to NijiV7.
@NowhereManGo thanks I'll check it out
This is an exquisite model, yet i consistently end with at least 7 fingers on each hand....
Edit - I have gotten it to work properly, and now am looking for a way to coax it to photo-realism.
I am purely looking at anime but i too noticed the 7 fingers. Were there any specific negative prompts you used to solve this?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.