User LoRA
XL
User LoRA
dnf,
sd1.5 v1早期模型，未经过融合或调试，效果较为一般，可以根据需求下载，有概率出现脸崩的情况，因此推荐使用Hires. fix，推荐分辨率7687682，训练和出图底模为anything-v5-PrtRE，sd1.5 v2加入了面部的训练数据，相较于v1面部效果有所改善但仍然推荐使用Hires. fix，加入面部数据外，还根据个人喜好融合了一些别的lora，进行了一定的改善，训练与出图底模为anything v5 prtresdxl版推荐权重在0.8，请勿在0.3-0.6的权重下出图，将会出现很奇怪的图，低于0.3则会出现另外一种很简洁的画风，xl没怎么玩过，这种情况还是第一次见，就很有趣，分辨率大于10241024即可，推荐11521152不推荐过度竖屏或者宽屏的尺寸，容易出现腰部很长的情况，训练底模与出图底模均为animagineXLV31_v31，本模型仅供娱乐，禁止任何商用与违法行为。pony版本使用大概340张dnf立绘所训练出的pony lora，训了好几天，这pony玩不明白，训练角色还好，训练画风让人头大大滴。找人练了一个版本，自己又训练了一个版本，总的来说效果还行，能用，提示词就dnf，pony lora权重一直玩不明白，预览图所用的是1，如果想要更加还原可以尝试增加到1.5，因此推荐权重0.8-1.5之间，请根据需求来调整，搭配角色lora，最好不要超过1，否则出图会很难看，出图的优化词是masterpiece,best quality,score_9,score_8_up,score_7_up,score_6_up，不确定有没有用，太玄学了v2版本将部分数据分开训练之后，融合了v1版本，整体风格相较于v1更加固定了，推荐权重0.6-0.8,如果觉得不够还原请尝试权重1，使用角色Lora则推荐0.4-0.6之间，尤其是以dnf角色为主的Lora，半身像不需要太大分辨率正常即可，但全身像请大于12801280The following uses Google Translate: sd1.5 v1 early model, not fused or debugged, the effect is relatively general, can be downloaded according to demand, there is a probability of face collapse, so it is recommended to use Hires. fix, recommended resolution 7687682, training and output base model is anything-v5-PrtRE, sd1.5 v2 adds facial training data, compared with v1 facial effect has improved but still recommend using Hires. fix, in addition to adding facial data, also according to personal preference to integrate some other lora, made some improvements, training and output base model is anything v5 prtre The recommended weight for the sdxl version is 0.8. Please do not use a weight of 0.3-0.6 to produce a very strange picture. If it is lower than 0.3, another very simple style will appear. I have not played xl much, and this is the first time I have seen this situation. It is very interesting. The resolution is greater than 10241024. 1152*1152 is recommended. It is not recommended to use a vertical screen or wide screen size, which is easy to cause a very long waist. The training base model and the output base model are both animaginXLV31_v31. This model is for entertainment only. Any commercial and illegal behavior is prohibited. The pony version uses about 340 dnf standing pictures to train pony lora. After training for several days, I don’t understand how to play this pony. The training character is okay, but the training style is very troublesome. I asked someone to train a version, and I trained another version myself. In general, the effect is OK and can be used. The prompt word is dnf. I have never understood the weight of pony lora. The preview image uses 1. If you want a more restored image, you can try to increase it to 1.5. Therefore, the recommended weight is between 0.8-1.5. Please adjust it according to your needs. With the character lora, it is best not to exceed 1, otherwise the picture will be ugly. The optimization words for the picture are masterpiece, best quality, score_9, score_8_up, score_7_up, score_6_up. I am not sure if it is useful. It is too metaphysical.After training some of the data separately in version v2, the v1 version is integrated, and the overall style is more fixed compared to v1. The recommended weight is 0.6-0.8. If you feel that it is not enough to restore, please try weight 1. When using character Lora, it is recommended to use between 0.4-0.6, especially for Lora mainly composed of DNF characters. The half body image does not need to have a large resolution, but the full body image should be larger than 1280 * 1280pixiv：05 - pixiv