Sign In

[Animagine]LimbusCompany Style

138
786
27.3k
53
Verified:
SafeTensor
Type
LoRA
Stats
539
26,292
7.4k
Reviews
Published
Apr 15, 2024
Base Model
SDXL 1.0
Hash
AutoV2
6414C4F379

相较于v1.0泛化性好了很多,某些概念过拟合的情况也得到了明显的改善,保留了底模构图的同时肢体控制和动作上会更稳定,在默认权重下风格会柔和一些,推荐控制在0.8-1.2之间来达到想要的画面效果。

如果人物在画面占比较少导致丢失细节,可以通过ADetailer来修复。

使用animagine角色列表中收录的触发词来还原人物,如:“1girl, ryoshu \(project moon\), limbus company”、“1girl, souryuu asuka langley, neon genesis evangelion”。

由于这个版本没有针对人物进行任何训练,完全依赖于animagine本身收录的角色,所以对于游戏内人物的身体比例、细节、还原度上不如上一个版本(比如良秀本来会随身携带的太刀),或许可以配合游戏内角色的lora来改善这个问题?

如果只是想还原游戏内立绘的服装、武器来达到类似假新闻的效果,可以使用1.0的版本,否则更推荐2.0的版本,无论是配合animagine收录的角色或者和其他的lora配合使用,都会有更好的效果。

如果对animagine的效果不满意,推荐尝试一下Ponymagine 9.1(质量提示词请使用animagine的版本),可能会有更好的表现。


gpt translate:

Compared to version 1.0, this version has much better generalization and has significantly improved the issue of overfitting certain concepts. While retaining the base model's composition, limb control and actions are more stable. The style is softer at the default weight, and it is recommended to adjust the weight between 0.8 and 1.2 to achieve the desired visual effect.

If the characters appear small in the image, causing a loss of detail, ADetailer can be used to fix this.

To recreate characters, use the trigger words included in the Animagine character list, such as: "1girl, ryoshu (project moon), limbus company" or "1girl, souryuu asuka langley, neon genesis evangelion"

Since no specific training was conducted for characters in this version and it completely relies on the characters included in Animagine, the body proportions, details and fidelity of in-game characters are not as good as the previous version (for example, Ryoshu originally carried a katana). Perhaps using LoRAs of in-game characters could improve this issue.

If you only want to recreate the costumes and weapons from in-game illustrations to achieve a fake news-like effect, you can use version 1.0. Otherwise, version 2.0 is recommended for better results, whether used with the characters included in Animagine or in conjunction with other LoRAs.

If you are not satisfied with the effects of Animagine, it is recommended to try Ponymagine 9.1 (for quality prompts, use the animagine version), which may perform better.