Detaylar
Öneriler
Noob-V-Pred-1.0 V1.1
Illustrious v1
Noob-V-Pred-1.0 V1
Epsilon-pred v0.1
V-Pred-0.65S v0.1 (old)
Dagasi Style for Noob/Illustrious

Dagasi Style for Noob/Illustrious

305
15
64
#Style
#รูปร่างมนุษย์
#เฟอร์รี
#น่ารัก
#ดากาซี
#Illustrious

Lora trained in the DAGASI's art style.

Use about 800 newer images for training, and manually proofread the prompt words, including furry and non-furry images. These images include some【限定】images. Unlike the first two parts of uncensored images that Dagasi will release, these images are bar censored, so I used PS to manually uncensor them. It maybe have an impact on training results. Because the training atlas is large enough, this lora can generate various characters such as furry, humans, feral, pony, etc. The coordination effect with other loras should be good and needs to be tested.

Dagasi's painting style is cute, and one of its characteristics is that almost every character has a blush on their face, so if you don't like blush, you can add it to the negative tags, but the effect will not be significant.

This lora also contains several Dagasi characters, as follows:

perro-kun:perro-kun, perro-kun \(dagasi\), fox boy, fox ears, fox tail, yellow eyes, two-tone fur, orange fur, brown fur, hair bobbles, brown hair,

yuki-chan:yuki-chan \(dagasi\), yuki \(dagasi\), leopard girl, leopard print, snow leopard, leopard ears, leopard tail, blue eyes, light blue hair, multicolored hair, multicolored fur, blue fur,

momo-chan:momo \(dagasi\), momo-chan \(dagasi\), fox girl, fox ears, fox tail, ears down, red eyes, gradient eyes, two-tone fur, brown fur, pink hair,

kohaku-chan:kohaku \(dagasi\), kohaku-chan \(dagasi\), tiger girl, tiger ears, tiger tail, tiger print, multicolored fur, tiger stripes, white fur, orange fur, black fur, green eyes, streaked hair, black hair, two-tone hair, orange hair, short hair,

buchi:buchi \(dagasi\), hyena girl, hyena ears, multicolored hair, pink hair, brown hair, multicolored fur, brown fur, blue eyes, hairclip, hair bobbles, futanari (if nsfw),

haruka:haruka \(dagasi\), rabbbit girl, rabbit ears, rabbit tail, brown hair, carrot hair ornament, red eyes, ears down, short hair, one side up, asymmetrical bangs, two-tone fur, brown fur,

In fact, I still don’t quite know how to set up the details of V-Pred lora training. At present, it seems that the Illustrious version has a higher fitting degree than the Noob V-pred version, which is reflected in the painting of human characters.

Weight: 0.6 - 1

Support the Artist: https://fantia.jp/fanclubs/34875

Enjoy.

使用Dagasi的画风训练的lora。

使用约800张较新的图片进行训练,手工校对tag。训练图像包含男性和女性furry以及少量非furry。在这些图像中包含了一些Dagasi发布的【限定】图像,不像Dagasi老师会发布的前两部分的无码图像,这些图像是有条形打码的,于是我用PS手工去码了,也许会对训练结果有影响。因为训练集足够大,该lora可以生成furry,人类,纯兽,小马等各种角色,和其他lora的配合效果应该也不错,有待测试。

Dagasi的画风偏可爱,其中一特点是几乎每个角色脸上都会有脸红,所以,如果你不喜欢常态的脸红,可以将该tag加入负面,但效果不显著。

这个模型还标注了几个Dagasi的角色,如下:

perro-kun:perro-kun, perro-kun \(dagasi\), fox boy, fox ears, fox tail, yellow eyes, two-tone fur, orange fur, brown fur, hair bobbles, brown hair,

yuki-chan:yuki-chan \(dagasi\), yuki \(dagasi\), leopard girl, leopard print, snow leopard, leopard ears, leopard tail, blue eyes, light blue hair, multicolored hair, multicolored fur, blue fur,

momo-chan:momo \(dagasi\), momo-chan \(dagasi\), fox girl, fox ears, fox tail, ears down, red eyes, gradient eyes, two-tone fur, brown fur, pink hair,

kohaku-chan:kohaku \(dagasi\), kohaku-chan \(dagasi\), tiger girl, tiger ears, tiger tail, tiger print, multicolored fur, tiger stripes, white fur, orange fur, black fur, green eyes, streaked hair, black hair, two-tone hair, orange hair, short hair,

buchi:buchi \(dagasi\), hyena girl, hyena ears, multicolored hair, pink hair, brown hair, multicolored fur, brown fur, blue eyes, hairclip, hair bobbles, futanari (if nsfw),

haruka:haruka \(dagasi\), rabbbit girl, rabbit ears, rabbit tail, brown hair, carrot hair ornament, red eyes, ears down, short hair, one side up, asymmetrical bangs, two-tone fur, brown fur,

事实上,我还是不太清楚V-Pred lora训练时的细节应该如何设置。目前看来,Illustrious版本的拟合度比Noob V-pred版本的更高,体现在对人类角色的绘画。

Weight: 0.6 - 1

支持画师:https://fantia.jp/fanclubs/34875

Çeviriyi görüntüle

Puan ve Yorumlar

-- /5
0 Puan

Henüz yeterli değerlendirme veya yorum alınmadı

F
Duyuru
2024-12-02
Model yayınlama
2025-03-05
Model bilgilerini güncelle
Model detayları
Tür
LORA
Yayınlanma tarihi
2025-02-09
Temel Model
NoobAI
Tetikleyici Kelime
dagasistyle
Kopyala
Sürüm tanıtımı

The training album is as of 【perro’s milking experience】

This version of lora is based on Noob-V-Pred-1.0. This version does not have to use Trigger Words. If you want to use it, use (dagasistyle:0.6~1.0).

This version is a further training of Noob-V-Pred V1.0. It slightly reduces the training set, modifies some training parameters, and trains about 50% more steps. Currently, this version can achieve better fitting.

(Note: This version is slightly overfitting. I think that for the painting style lora, a slight overfitting is desirable. Please reduce the weight appropriately when using it. About 0.7~0.8 is the best, and use detailed tags.)

(All sample images use Noob-V-Pred-1.0 and are generated using ComfyUI. V-Pred-1.0 cannot be selected when uploading images.) (Note: In the example picture, the weights of the two quality-improving loras are both lower, 0.35)

Future updates: Training will be carried out after Mr. Dagasi updates a certain amount. Some of the training images used in this lora were tagged earlier. I was not familiar with lora training at that time, so I will re-check the tags in my spare time.

训练图集截止至【perro的挤奶体验】。

该版本lora基于Noob-V-Pred-1.0。该版本不必须使用Trigger Words,如果要用,则用(dagasistyle:0.6~1.0)。

该版本是Noob-V-Pred V1.0的进一步训练,稍微减少了训练集,修改了部分训练参数,多训练了约50%的步数,目前该版本能够达到更好的拟合。

(注:该版本略微过拟合,我认为对于画风lora,稍微过拟合是可取的,在使用时请将weight适量降低,约0.7~0.8是最好的,并且使用详细的tags。)

(示例图像全都用Noob-V-Pred-1.0,使用ComfyUI生成,上传图片的时候选不了V-Pred-1.0,到现在也只有0.75的选项,不知道什么时候能修一修。)(注意:示例图片中,两个提升质量的lora的重量都是较低的,为0.35)

今后的更新:会在Dagasi老师更新一定量后再进行训练,本lora使用的训练图集中有一部分是在较早之前标注tag的,当时我还不熟悉lora训练,我会在空余时间重新校对tag。

Lisans Kapsamı
Model Source: civitai

1. Yeniden paylaşılan modellerin hakları orijinal yaratıcılara aittir.

2. Orijinal model yaratıcıları, modellerini talep etmek için resmi kanallar aracılığıyla SeaArt AI personeliyle iletişime geçmelidir. Talep etmek için tıklayın

Yaratım Lisansı Kapsamı
Çevrimiçi Görsel Üretimi
Birleştirme İşlemi
İndirime İzin Ver
Ticari Lisans
Üretilen resimler ticari amaçlarla kullanılabilir veya satılabilir
Modelin yeniden satılmasına veya birleştirildikten sonra satılmasına izin verilir
QR Code
SeaArt Uygulamasını İndir
Mobil cihazınızda AI yaratıcılık yolculuğunuza devam edin