Higurashi no Naku Koro ni Kira Style: https://civarchive.com/models/154232
When using the Civitai Generator, if you do not enable Enhanced Compatibility, abnormal results will be generated!
Setting the weights to approximately 1.0 is recommended.
Please Use Anima Checkpoint!
Activation tag : @higurashi_mei
Anima Recommend Prompt
Positive: This anime illustration has excellent quality, and is a clean drawing hand-drawn by a human. year 2025, Exceptional quality that stands out as a truly outstanding illustration. (score_9, score_8, score_7:1.2), (masterpiece, best quality, good quality, highres, absurdres:1.2), @higurashi_mei, (tag),Negative: This anime illustration is unnatural and strange, and is an ugly and subpar illustration. worst quality, (worst aesthetic:1.1), bad quality, score_1, score_2, score_3, score_4, score_5, lowres, bad anatomy, (artistic error:1.2), (bad:1.2), off-topic, multiple views, comic, extra digits, fewer digits, fewer, error, missing, jpeg artifacts, artist name, signature, twitter username, username, logo, watermark, scan, unfinished, variations, bad hands, NoobAI Recommend Prompt
If the generated image looks strange, set the CFG Scale to between 5 and 7.
Recommend Prompt
Positive: higurashimei, (tag), (masterpiece, best quality, absurdres, highres), (newest, very awa), Negative: worst quality, worst aesthetic, bad quality, artistic error, old, ealry, mammal, anthro, furry, feral, semi-anthro, lowres, (bad), off-topic, multiple views, comic, very displeasing, displeasing, chromatic aberration, (extra digits), fewer digits, fewer, text, error, missing, jpeg artifacts, artist name, signature, twitter username, username, watermark, scan, unfinished, [abstract], variations, (((bad hands))), ai-generated, Characters supported by LoRA:
Furude Rika
Hanyuu
Houjou Satoko
Ryuuguu Rena
Houtani Nao
Maebara Keiichi
Sonozaki Mion
Sonozaki Shion
Akasaka Mamoru(It comes out a bit weird)Akasaka Miyuki
Kimiyoshi Kazuho
Tamurahime No Mikoto
Tomitake Jirou(It comes out a bit weird)Takano Miyo
The other characters may be able to be output.
Description
I switched to V-Pred, added an additional dataset, slightly modified the training method, and determined that this version achieves the highest recall.
FAQ
Comments (4)
Munsy, can you change the SDXL base model tag from the Epsilon versions to Illustrious and leave this new V-Pred 1.0 version as the SDXL one?
I'm asking this because there are more checkpoint options for generation on the Illustrious tag and the epsilon lora works really well with them, but not that good with the V-pred checkpoint.
If one day they also make the epsilon Noob checkpoint available for generation, you can change it back again.
If not possible, it's 100000% completely fine. Also I just want to say again that I'm grateful for all your releases and updates. They always look great!
Munsy, thank you very much for everything. Wish you a great Christmas! Have a nice day!
I've fulfilled your request; check out DoRA NoobAIXL11 v2.0!
I'll change it back once most of the NoobAI checkpoints become available for use as generators again.
@munsy0227 Perfect! Thank you very much, Munsy!!!