lora有水印，所以反向提示词里记得添加“水印”。这个lora里有面部，所以与其他lora混着使用时，最好使用lora分层插件，看看lora在不同权重下的表现<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6da90d73-945e-4d05-afec-69cfededc40e/width=525/6da90d73-945e-4d05-afec-69cfededc40e.jpeg" />当步数在28时，推荐权重0.4～1本人电脑性能有限，大家可自行测试。lora训练基础参数大模型：chilloutmixvae:vae-ft-mse-840000-ema-pruned采样方法：DDIMThere is a watermark in the Lora, so remember to add the word "watermark" in the reverse prompt when using it with other Loras. As this Lora contains a face, it is recommended to use a layered Lora plugin when mixing it with other Loras to see how Lora performs at different weights. When the step number is 28, a weight of 0.4 to 1 is recommended. Please note that my computer performance is limited, so you can test it yourself. The basic training parameters for Lora are as follows:Large model: chilloutmixVAE: vae-ft-mse-840000-ema-prunedSampling method: DDIM.

Files with SHA256: 328b2c64dc2ab13c81d9f42c8e3a2b46b7d504ee26c297030d19fdb5cef1cee6