Yuzuriha | Jigokuraku (Hell's Paradise)
Results:
Overall score :1/7 Better than nothing.
Character settings: 1/2
Without the other prompts recommended by the author, some details are missing. It is suspected that the tag is added too much during training, and if you want to perfectly restore the character, you need to add some more accurate descriptions, and there is a risk of confusion with other concepts in use.
Style: 0/2
Except for the lofi model, all of them generate 2D styles, and it is suspected that only animation screenshots are used during training, and the animation style is not eliminated, resulting in too large weight of this style, and it is difficult to directly generate 3D images.
Failure: 0/2
The presence of multiple people and gender errors, as well as the background being too monotonous (forest), can be attributed to deficiencies in the material and training parameters.
Subjective: 0/1
The model gets the most 2D detail bonuses in Nightsky, generating 3D images but missing more detail. At the moment it is only a half-finished product, expect better LoRa.
Notes:
This test is intended to indicate the availability of the character LoRa. 40 images were generated using four seeds and ten models. Evaluate from three dimensions of character setting, style and failure, and one subjective dimension.Objective dimension score: 0 not recommended, 1 available, 2 recommended, subjective dimension: 0 not recommended, 1 recommended, out of 7 points.
结果：
总分：1/7 总比没有好。
人物设定：1/2
如果没有作者推荐的其他提示，一些细节就缺失了。怀疑标签在训练时添加太多，想要完美还原角色，需要添加一些更准确的描述，并且存在与使用中的其他概念混淆的风险。
风格： 0/2
除了lofi模型，都是生成2D样式，并且怀疑在训练时只使用了动画截图，并且没有消除动画样式，导致这种样式的权重过大，很难直接生成3D图像。
故障：0/2
多人的存在和性别错误，以及背景过于单调（森林），可归因于材料和培训参数的不足。
主观：0/1
该模型在Nightsky中获得了最多的2D细节奖励，生成了3D图像，但缺少更多细节。目前它只是一个半成品， 期待更好的 LoRa.
注意：
该测试旨在指示字符 LoRa 的可用性.使用四个种子和十个模型生成了40个图像。从人物设定、风格与失败三个维度和一个主观维度进行评估。客观维度得分：0 不推荐，1 可用，2 推荐，主观维度：0 不推荐，1 推荐，满分 7 分。