How to keep structure and create more diversities using img2img

[pujunting]

thanks for your excellent job. I trained my own model,then i use img2img,my target is create more diversities and maintain structural informations of init-img as much as possible. Is there only the "strength" parameter can control it?Or something else I can do,to detach the structure from conditional image,then apply to a new distribution? hope for answer

[yanjingke]

请问您怎么训练的

[eslambakr]

Can u let us know how did u train it?