@darkhawke
I use several techniques. First I try to get the result I have in mind by just prompting. Who knows, might get lucky. If that doesn’t work (aka after around 100 pics), I generate pony and background separately, put them together manually and let the AI inpaint the result to make it look seamless. Last resort is I sketch out the picture myself and inpaint that. The last two give the best result but are more work of course. And if that is done, I have my 720p starting point. Then I edit & inpaint and all errors and stuff I want changed, clean it up, upscale to 2k and clean again. (Since details can get mushy in the upscale process.)
Then I sit several days on the picture to make sure I didn’t miss anything (you get “blind for errors” after working to long on one picture. Trick is to “flip” it and if it still looks decent, you got most of it.). If I still like the result, then I upload.
If you want to talk more about stuff like this I would recommend you join the Tantabus Discord. I am on that server as are others who can tell you more about their processes as well.
@truekry
Yeah, I was more thinking of in-painting. Just generating images is dead simple, but creating images as consistent and well done as you do seems significantly more complicated.
Do you mock out a pose first? Gen a bunch of starter images till you get one close and then try to tweak it? Stuff like that.
@tyto4tme4l
I think any good picture drawing or editing software has a variant of this tool today. The point of showing it in photopea is that anyone, without installing any of them, can make this quick edit in under a minute on any computer with a browser. Something you would need to visit this site anyway.
@Phemto
It works on a lot of stuff. The Band Aid tool is semi AI. It looks at the surrounding pixels and tries to guess what should be there. If nothing else, it is a great base for larger edits.
@darkhawke
I could make one for the program I use, but there are so many and most are one click install, download a model, put in what you want and let it rip. There is not really much too it anymore. Inpainting works a little different depending on the program used (cause of UI). But normally you designate an area in a pic with a mask and let the AI try to regenerate that area.
Erasing the ghosts’ signatures is a great way to get yourself haunted!
But seriously, I had no idea about this. I’ll try to keep in in mind where there’s a small thingy (I’m assuming it works on more than just sigs) that I want to get out of an image.
I use several techniques. First I try to get the result I have in mind by just prompting. Who knows, might get lucky. If that doesn’t work (aka after around 100 pics), I generate pony and background separately, put them together manually and let the AI inpaint the result to make it look seamless. Last resort is I sketch out the picture myself and inpaint that. The last two give the best result but are more work of course. And if that is done, I have my 720p starting point. Then I edit & inpaint and all errors and stuff I want changed, clean it up, upscale to 2k and clean again. (Since details can get mushy in the upscale process.)
Yeah, I was more thinking of in-painting. Just generating images is dead simple, but creating images as consistent and well done as you do seems significantly more complicated.
I think any good picture drawing or editing software has a variant of this tool today. The point of showing it in photopea is that anyone, without installing any of them, can make this quick edit in under a minute on any computer with a browser. Something you would need to visit this site anyway.
It works on a lot of stuff. The Band Aid tool is semi AI. It looks at the surrounding pixels and tries to guess what should be there. If nothing else, it is a great base for larger edits.
I could make one for the program I use, but there are so many and most are one click install, download a model, put in what you want and let it rip. There is not really much too it anymore. Inpainting works a little different depending on the program used (cause of UI). But normally you designate an area in a pic with a mask and let the AI try to regenerate that area.
Edited