测试完成了。接下来我会弄2.0版本的，全新的训练集处理和新的参数
The test is complete. Next, I will work on version 2.0, new training set processing, and new parameters
更正一下，trigger name 是我乱填的 a woman
真正的tirgger name 是embedding本身的名字 我已经更正了。为这个失误抱歉
Correct it, the trigger name is a woman that I filled in incorrectly
The real tiger name is the name of embedding itself. I have corrected it. Sorry for this mistake
Don't underestimate emb by it's only over a dozen kb. Although this metaphor is not appropriate, early computers were as big as houses, but how big are chips today?
In a way embedding is better than lora with face loyalty restoration, hope more bro try to train your emb.
Description
The problem of facial collapse at long distances has not been resolved, but the direction has been found. I will try to improve it in subsequent versions and other works. The preview image is rendered by SD 1.5. If it is a different base model, it should have a different (better?) performance