出于将AI融入我的资产制作流程，我制作了这个Lora，基于SD2.1，50张512*512图像训练集，总步数10000步，这是我第一次制作lora模型，没想到效果还不错，该模型有一些过拟合的问题，所以使用时请把参数调到0.6以下. &lt;lora:Bonsaitree:0.5&gt;实例图大模型使用的是<a target="_blank" rel="ugc" href="https://civarchive.com/models/22922/lyriel">https://civarchive.com/models/22922/lyriel</a>（PS:我明明训练是基于SD2.1，但这个Lora却不能用2.0以上的大模型来出图，只能使用1.5的模型，希望有大佬能给予一些解答，谢谢）示例图是文生图直出，如果想要效果更好，可以：1.通过二次元模型生成大造型2.通过图生图+tiled diffusion运用该Lora补充细节，转为三次元（还有好多问题，想找一找组织，Q362988774）For integrate AI into my asset production process, I created this Lora,Based on SD2.1, 50*512*512 images, total steps 10000,This is the first time for me to make lora. Unexpectedly, the effect is quite good. The model has some overfitting problems, so please adjust the parameter below 0.6 when using.&lt;lora:Bonsaitree:0.5&gt;Instance figure model used: <a target="_blank" rel="ugc" href="https://civarchive.com/models/22922/lyriel">https://civarchive.com/models/22922/lyriel</a>(PS: I trained based on SD2.1, but this Lora can't use models above 2.0 to draw pictures, only use 1.5 models, hope someone can give some answers, thank you)The sample diagram is a tex2img straight out. If you want better results, you can:1. Generate large shapes through 2D model in tex2img,2. Generat details through img2img+tiled diffusion in 3D model.

Files with SHA256: 77fe241df51a8bc67c90d07c4d4fd7c28636075c4df1996143620ca4a46f4b23