Loading
Du hast einen ComfyUI-Workflow? Erstelle direkt im Browser eine verständliche Workflow-Dokumentation.Workflow verständlich machen.Jetzt ausprobieren ↗
This LoRA allows you to adjust the overall level of detail by modifying its weights, with a particularly noticeable impact on the background. This LoRA is versatile, suitable for various 2D, 2.5D, and 3D SDXL models.
Modellquelle
Quellenbeschreibung
这是一个可以通过改变权重来调节画面整体细节的LoRA,其对背景的作用尤为明显。该LoRA适用于各类2D,2.5D和3D模型。
你也可以通过将其以特定权重融合到大模型中,从而获得细节更丰富、背景更充实,或,画风更简化、线条更简洁的风格的模型。
This LoRA allows you to adjust the overall level of detail by modifying its weights, with a particularly noticeable impact on the background. This LoRA is versatile, suitable for various 2D, 2.5D, and 3D SDXL models.
You can also merge it with specific weights into a base model to achieve different styles, such as more intricate backgrounds and richer details, or simplified and cleaner lines.
使用建议 Usage Guideline
在负值权重下,背景的近景会显著变清晰,远处细节如楼房等立面的复杂度会增加,人物头发及衣服线条细节增加。在正值权重下,背景变得更模糊,人物的线条细节也会得到简化。需注意,对于一些2D模型,由于线条和色块的复杂度也被用于区分不同的风格,所以当你在使用这个LoRA改变细节时,也会对2D画风有一定影响。
Quellen
1 QuelleModellartefakte
1 ArtefaktQuellenauszüge
1 AuszugSource context: 6339 downloads · Type LORA · Base model SDXL 1.0
When using negative weights, the foreground of the background becomes significantly clearer, and distant elements like building facades gain complexity. Moreover, finer details in characters' hair and clothing will be enhanced. Conversely, positive weights lead to a more blurred background and simplified lines in characters. Note that when applying this LoRA to some 2D models, as lines and color blocks are also used to differentiate styles, altering the details with LoRA might impact the 2D style to some extent.
权重取值建议:建议先从-1到1进行尝试,再根据效果进行调整。根据我对于目前的SDXL模型的测试经验,如果你使用拟合程度较高(过拟合)的base模型,那么建议保持此LoRA权重在-1到1之间。而对于一些拟合程度良好,泛化性优异的模型,则可以在权重-2到2之间使用此LoRA.
Suggested weight values: It is recommended to start with weights ranging from -1 to 1 and then adjust according to the desired effects. Based on my testing experience with the current SDXL models, if you are using a base model with a high degree of fitting (overfitting), it is advisable to keep the LoRA weights within the range of -1 to 1. For models with excellent generalization and fitting, you may experiment with LoRA weights between -2 and 2.
如果你喜欢我的模型,请不要吝啬你的点赞啦!模型定制或交流学习,欢迎添加我的微信:Anson_NgAu,来支持我,与我一起讨论吧!
If you like my model, please don't hesitate to give it a thumbs-up! For model customization or to exchange ideas, feel free to add me on WeChat: Anson_NgAu. Your support and collaboration are highly welcome!
注 Note
您不得将此模型及其衍生版本(如融合模型版本)托管于计划赚取收入或捐赠的网站/应用程序。如果您需要将此模型及其衍生版本用于商业目的(如生成式服务、售卖图片、将图片用于公开发表的文章或出版物等),请通过微信(Anson_NgAu)联系我。
You are not permitted to host this model or its merged versions on websites/applications that generate income or accept donations. If you intend to use this model or its merged versions for commercial purposes (such as generative services, selling generated images, using images for publicly published articles or publications, etc.), please contact me via WeChat (Anson_NgAu).
您不得直接售卖由此模型生成的图片,除非您对该图片进行了足够程度的人工修改,使其在法律意义上可以被完全判定为您的个人作品。如果您违反本条,所造成的一切法律后果由您个人承担,请恕本人概不负责。
You may not directly sell images generated by this model unless you have made sufficient manual modifications to the extent that the image can be legally considered your own work. If you violate this provision, you will be solely responsible for any legal consequences, and I cannot be held liable.
您可以将此模型及其衍生版本、通过此模型生成的图片自由用于免费/公益用途,包括免费的图片生成式服务。
You are free to use this model and its merged versions, as well as images generated by this model, for non-profit/charitable purposes, including free image generative services.
您可以自由融合此模型,如果您发布此模型的融合模型,还请您标注使用了此模型并附带链接,十分感谢!
You are allowed to merge this model freely. If you publish a merged model based on this model, please credit the use of this model and add a backlink to this page. Thank you very much!
您不能使用该模型故意制作或共享非法或有害的输出或内容,请您遵守公序良德,将此模型用于积极正面的用途。
You are prohibited from using this model intentionally to create or share illegal or harmful output or content. Please adhere to moral and ethical standards and use this model for positive and constructive purposes.
