The model was trained with 1288 images and could generate T-shirt images.
size: 512×512
The two-dimensional performance is good, and it can generate character prints, but the realistic style is not good, and it is better to generate no character prints.
You can try other bottom film plots, I haven’t used them all.
Color contamination may occur because the number of render steps is insufficient, so increase the number of iteration steps.
Just enter the t-shirt and you can summon it, that is, a T-shirt with a random color print pattern will be generated.
Otherwise, you can describe exactly, if you want to change the print, you can add a prompt×××× print,
For example, fire print,
The model does not generate a good print of the character and needs to be gachaked.
I’d rather have the character in the print than in a T-shirt.
So, if you don’t want characters, you can add no humans to the positive prompt
Negative prompt words plus people. Since many of the bottom mask characters are very strong in nature, there is still a possibility that characters may appear.
And no humans will affect the character print, and it will think that the print of the clothes is also a person.
You can debug it.
If you use it with LoRa, it’s not good and easy to break.
You can also try, if you can debug the parameter weights can be matched with the character lora to make it a print, welcome to let me know.
If it is a light-tempered lora, you may not need to reduce the weight, just use it.
You may need to add print to the character’s summon word, and the weight of the larger LoRa needs to be lowered.
If you like it, please like it. I’ll probably update and find a way to match the character LoRa in the T-shirt print.
Welcome to use, comment and return images.