Detalhes
Recomendado
v1.0
v0.25
v0.5
v0.0
Uruha rushia潤羽るしあテストLoRA 960p

Uruha rushia潤羽るしあテストLoRA 960p

1.3K
98
320
#Personagem
#Vtuber
#hololive

Basically, it is preferable to use v0.0 when considering using it with other LoRA.

v0.5 and v1.0 are quite over-learned in order to test whether it is possible to output multiple types of costumes by adjusting the weight of the number of learning iterations.

基本的に他のLoRAとの併用を考える場合、v0.0を使用することが望ましいです。

v0.5とv1.0は複数種類の衣装を、学習繰り返し回数の比重を調整すれば出力できないかを試すためにかなり過学習気味になっています。

v0.25 (epoch2) is a model that I uploaded only for the purpose of a comparison test to see if multiple costumes can be produced accurately as the number of epochs progresses.

v0.25 (epoch2)はあくまでもepoch数が進むほど複数の衣装は正確に出せるようになるのかを比較テストするためだけにアップロードしたモデルです。

v0.0 is a re-upload.

v0.0は再アップロードです。

v0.0 is trained on 300 768p images

v0.0は300枚の768pの画像でトレーニングされています

First of all, please understand that this is not very good LoRA.

It's not LoRA that can output multiple costumes.

LoRA can only output Rushi uruha's initial costume, but it is a model that has been verified to increase its expressiveness to the limit.

As a verification, in addition to Rushia uruha's initial hairstyle, I tried to mix data sets of multiple costumes, but the number ratio was less than 1:5, so it was almost impossible to output other costumes. think about it

Also, a known bug is that typing detached collar will almost certainly force the ????? into forced focus.

The reason for this is that I failed to adjust the ratio of the number of sheets because I tried to learn the shape of the collar properly.

A good sampler would be dpm++ 2m karras or DDIP.

まず初めに、これはあまりいいLoRAではないことを理解してください。

複数の衣装を出力することができるLoRAというわけでもないです。

ただただ潤羽るしあの初期衣装しか出力できないLoRAですが、極限まで表現力を高められないかと検証したモデルです。

検証としては初期衣装初期髪型の潤羽るしあに加えて複数の衣装のデータセットを混入させてみたものの、枚数比率が1:5以下となってしまったため、ほかの衣装はほぼ出力できないと考えてください。

また、既知のバグとして、detached collarと入力した際にはほぼ確実に胸が強制フォーカスとなってしまいます。

これの原因は襟の形状をきちんと学習させようとしたがために枚数比率の調整をしくじったためです。

サンプラーはdpm++ 2m karrasもしくはDDIPがいいと思われます。

v0.5 and v1.0 were trained on 4500 training data, 20000 regularized images and 1000 other costumes.

v0.5とv1.0は4500枚の教師データと、20000枚の正則化画像とその他の衣装1000枚でトレーニングされています。

LoRA posted in the past can be downloaded at HuggingFace.

過去に投稿したLoRAはHuggingFaceでならばダウンロードできます。

https://huggingface.co/Lineins/All_Uruha_Rushia_LoRA

Ver tradução

Classificação e Avaliação

-- /5
0 classificação(ções)

Sem classificações ou avaliações suficientes

no-data
Sem dados
avatar
Conversar com o modelo
Anúncio
2023-07-10
Publicar Modelo
2024-04-15
Atualizar info do modelo
Detalhes do modelo
Tipo
LORA
Tempo de Publicação
2023-07-10
Modelo básico
SD 1.5
Palavra de ativação
<lora:testbluerushia2_3:1>, bluerushia, green hair, double bun, red eyes, short hair, blue bow, hair ornament, blue dress, detached sleeves,
Copiar
Introdução à Versão

First of all, please understand that this is not very good LoRA.

It's not LoRA that can output multiple costumes.

LoRA can only output Rushi uruha's initial costume, but it is a model that has been verified to increase its expressiveness to the limit.

As a verification, in addition to Rushia uruha's initial hairstyle, I tried to mix data sets of multiple costumes, but the number ratio was less than 1:5, so it was almost impossible to output other costumes. think about it

Also, a known bug is that typing detached collar will almost certainly force the ????? into forced focus.

The reason for this is that I failed to adjust the ratio of the number of sheets because I tried to learn the shape of the collar properly.

A good sampler would be dpm++ 2m karras or DDIP.

まず初めに、これはあまりいいLoRAではないことを理解してください。

複数の衣装を出力することができるLoRAというわけでもないです。

ただただ潤羽るしあの初期衣装しか出力できないLoRAですが、極限まで表現力を高められないかと検証したモデルです。

検証としては初期衣装初期髪型の潤羽るしあに加えて複数の衣装のデータセットを混入させてみたものの、枚数比率が1:5以下となってしまったため、ほかの衣装はほぼ出力できないと考えてください。

また、既知のバグとして、detached collarと入力した際にはほぼ確実に胸が強制フォーカスとなってしまいます。

これの原因は襟の形状をきちんと学習させようとしたがために枚数比率の調整をしくじったためです。

サンプラーはdpm++ 2m karrasもしくはDDIPがいいと思われます。

v0.5 and v1.0 were trained on 4500 training data, 20000 regularized images and 1000 other costumes.

v0.5とv1.0は4500枚の教師データと、20000枚の正則化画像とその他の衣装1000枚でトレーニングされています。

Escopo da licença
Fonte: civitai

1. Os direitos dos modelos repostados pertencem aos criadores originais.

2. Criadores originais que desejam reivindicar seus modelos devem contatar a equipe do SeaArt AI pelos canais oficiais. Clique para reivindicar

Escopo da licença de criação
Transmissão ao vivo
Fusão
Permitir download
Licença comercial
Imagens geradas podem ser vendidas ou usadas para fins comerciais
Permitir revenda de modelos ou sua venda após a combinação
QR Code
Baixar App SeaArt
Continue sua jornada de criação com IA no celular