Problems at the training with new models

[nativenoble]

Good morning everyone,
I have a problem of understanding, last week I had my pictures trained against the samdoesart model and the results were really great. Now I have tried it against 2 other models and the results are absolutely not good.
Against the samdoesart model I also have to use my trigger word that my images are taken. In the 2 other attempts, my image is always taken no matter if I use the trigger or not. For example, if I say take a selfie with Bruce Willis I have my face in the picture 2 times. What am I currently doing wrong? Has something changed?

[chewtoys]

i think you set Text_Encoder_Training_Steps too high, start like with 250, train the model, check it, if its too low add another 50 and so on.
make sure UNet_Training_Steps is set to 0

[nativenoble]

thanks for your reply, I have selected the default setting and that is 250 steps, should I set it even lower?

[chewtoys]

no idea im just guessing, the scrip was changed yeah, lots of people having problems

I did follow this advice, and it seems to be working
#1162 (reply in thread)

To get a good balance between training time and good results, you need to do some tuning, start with 600 unet at 2e-5, if not enough, resume the training for 200 steps with a slightly lower lr 8e-6 until you get good results, the text encoder should not go over a total of 350 for 10 images during the whole training.

650 2e5 - looks terrible
250
then
150 8e6 - looks way better
0

repeat 8e6 until it look like the model you want, cfg scale 12 to test