✦ This month REWARDS (Part 2). ART + LoRA (Ying – Horse of High Peaks)
V6.0
In this update, I have incorporated real human texture details to make the generated images more realistic. The model still supports multiple styles and fully inherits all data from Version 5.0, including all artist-specific styles. When using it, you can input an artist’s name (e.g., NIXEU, wlop, quasarcake, sakimichan). This will add an extra layer of the artist’s signature style to Version 6.0’s base model, creating a deeper artistic atmosphere.
If the skin texture does not meet your expectations during use, try entering "realistic" or other valid texture-related prompts. This update also aims to optimize the performance of my character LoRAs—most of the test images you see were created using my original LoRA models. Due to special reasons (e.g., being custom-made or exclusive to private use), these LoRAs cannot be publicly shared on CivitAI for download. However, I have uploaded them to Patreon and AZZ, where you can access all exclusive models. I update at least 3–4 exclusive models every month.
My goal in refining Version 6.0 is to better serve my clients and make their custom LoRAs look more polished. If you are satisfied with my work, please follow me for more updates. Thank you!
artist:NIXEU
artist:wlop
artist:quasarcake
artist:Ilya Kuvshinov
artist:Rolua
artist:toi8
artist:sakimichan
artist:yoneyama mai
artist: Jada Bruney
artist: Jarom Vogel
artist: Ilya Kuvshinov
artist: Jesper Ejsing
artist: Sam Yang
artist: sakimichan
artist: Laserflip
artist: Khyle
artist: 92M
artist: hews
artist: Cutesexyrobutts
artist: LiangXing
artist: Machi
artist: Takeshi Obata
-------------------
You're only seeing a fraction of the artists here, but you can still try looking up some of the more famous artists on X. Using them with V6.0 can give their styles a different kind of flair.V5.0
In this update, the model has been embedded with the VAE that is most suitable for it. Please set the VAE to "None" when using it.
This update fixes better hand details, supports image generation with all well-known artist keywords and different artistic styles, and is compatible with various Illustrious LoRA models.
This update has integrated all the data from my previous version of the model (Hs Art Anime Godlike + Hs UltraHD CG [ill]Epic).
This model is the most, most, most suitable checkpoint for all my 3D character LoRAs. If you want to use my 3D character LoRAs, I do not recommend you use checkpoints from other authors, as the results may not be ideal.
This update is also to better adapt to all my 3D Chinese animation character LoRAs.
I will train Chinese animation character LoRAs in the coming time. All character LoRAs are custom-made. If you need customization, please send me a private message.
This update may be the final one. If no major issues are encountered, I will no longer update this checkpoint because it almost perfectly meets all my creative needs.
If you encounter problems while using it, it must be your own issue. Please do not come to blame me; I am not responsible for teaching you how to use it!
If you need me to teach you, my time comes with a cost. Thank you for your understanding.
如果您是中国用户，下载速度很慢，或者不想逐个 下载。我将这一年内创作的模型打包成了合集， 放在了我的网盘中。
合集链接：https://pan.quark.cn/s/18999506dfaa
Dedicated Post for Custom Models
Illustrious Version
About this version
Recommended Settings
Parameters
CFG: 5–7.5(may vary based on use case)
Steps: 20–28
Sampling Method: Euler a
Prompt Prefix:
masterpiece, best quality, high quality, detailed, absurdresNegative Prompt:
ugly, bad feet, bad hands, bad art, ugly artstyle, bad anatomy, bad fingers, censor, censored, ugly, deformed, noisy, blurry, low contrast, photo, realistic ,watermark, username, textPony Version
Recommended settings:
Sampling methods: Euler a
Negative prompts are not necessary.
CFG: 2 - 9
For full - body portraits, it is recommended to use the ADetailer extension to enhance facial details.
Positive Prompt
score_9, score_8_up, score_7_up,Negative Prompt
score_6,score_5,score_4,Description
FAQ
Comments (20)
Great model, very responsive, actually delivers on 3DCG look
Appreciate your positive comment. Glad it meets your expectations.
Really loving the consistent CG texture! But one question, does this model support all innate IP characters present in pony? and also I noticed some different writing structure for IP characters [eg: makima \(chainsaw man\)] than Pony. Can you tell me how to write other characters?
I haven't really paid much attention to the format when writing IP characters. You know, as long as it's some globally renowned IPs, AI can recognize them. Even if you just input the name, AI will know who you're referring to. If you want to generate some artistic images of characters, I'd highly recommend using the Illustrious model. It has a better understanding of most character IPs.
@HS_hi Thank you! I don't know how to use Illustrious ckpt in general as of yet but when I do I'd def try your model and again really loving this model of yours <3 Cheers!
@annikaherzfeld Thank you very much. You can run it using webui_forge or ComfyUI.
in my experience it helps to download "a1111-sd-webui-tagcomplete"
as you type a name it will give you the most common tags used on whatever booru site it collects from.
correctly describing traits can also help nudge pony or ill in the right direction (red hair, braided ponytail, makima \(chainsaw man\)
from what I understand illustrious is biased towards mostly Asian media which steers it away from Western media.
pony also had a year of different people adding in their own datasets of characters that most likely were not included for illustrious with exceptions with models that were remade like hassaku.
Also, be sure to use the quality tags in this models description.
@HS_hi I'm trying diff illustrious models now and also gave your latest Ill. model a run. It's producing weirdly high contrast images, tho my other ill models are working fine. Also imo the pony version of your model has a better texture render in terms of consistency along generations. wdyt?
@annikaherzfeld When you notice that the images you generate are vastly different from others', you can download an image created by someone else and use the same parameters for generation. If the results are completely different, there must be an issue with your tool. Since I don't have access to the tool you're using, I can't determine where the problem lies. Judging from what you've described, it's possible that the model wasn't fully loaded or there are some problems with the webui settings that are causing this.
Awesome work! Can you please create a LORA version of this model (like lora extract)?
I'm really sorry, but I have no intention of doing that. All the models I create are mainly for my own convenience. This includes all the LoRAs I make. I only upload them to prevent local file loss.
@HS_hi If you don’t mind, could you share how you train or merge models? The quality of your results clearly shows that you know exactly what you're doing. Any documentation or resource recommendations would be greatly appreciated!
I'm currently trying to achieve a specific semi-real look with Pony/Derivatives, but I haven't quite nailed it yet :'). Your model is truly impressive in terms of quality and consistency—character features remain remarkably intact across generations, and the level of detail with almost zero artifacts is incredible. My favorite part is how it renders hair—it’s the best I’ve gotten so far! I’m really looking forward to seeing what other gems you release in the future.
P.S. If you're comfortable with it, I'd love to share some of my own training results/images and get your experienced and valuable feedback on how I can improve and better achieve my vision. But no worries if that's not possible! Cheers, and thanks again for the amazing model. :)
I'm sorry to disappoint you. I'm in a tight spot in my life at the moment, so I'm probably in a low mood and not eager to communicate much with others. Moreover, I don't have extremely high - level skills. I'm just a bit more industrious than ordinary people. I use Photoshop to cut out each image, removing the content that I don't want the AI to learn. I manually edit the labels of each picture required for training. After the training is completed, I'll conduct a test. If I'm not satisfied, I can only adjust parameters such as the learning rate and the number of training steps and then carry out another round of training. I hope this will be helpful to you.@jokeerboi629
@HS_hi I pray that things get better for you soon. Appreciate you replying :)
This model is amazing at following a string of detailed prompts and it doesn't just do lewd girls. It does guys well too. Brilliant! Thank you for this.
Thank you for your recognition.