2D 3D-Blender Fan Art Some Fan Arts I Made (NSFW)(HOT)

Jul 26, 2022
32
17
Pink Helmet girl is probably my choice, they seem to have more detail as they go down.

What is the highest native render resolution you can do for me its around 1000x1000 but that will go from 2-4 it/per second to 30-60 seconds per it

Also are you using the same seed to img to img upscale, or just ERSGAN
No I am not using the same seed but I am using img to img upscaler and I am Using ESRGAN and CodeFormers after img to img
As for native render i can do more than 1000x1000 but it just consumed more time.
Some Times I use Photoshop to add more details.
 
Jul 26, 2022
32
17
Pink Helmet girl is probably my choice, they seem to have more detail as they go down.

What is the highest native render resolution you can do for me its around 1000x1000 but that will go from 2-4 it/per second to 30-60 seconds per it

Also are you using the same seed to img to img upscale, or just ERSGAN
One More If you don't Mind from where you get LORAS ?
 

felldude

Member
Aug 26, 2017
448
1,399
One More If you don't Mind from where you get LORAS ?
Civitai probably has the most, I train my own but I have used a few other people have made also...I dont have the power needed to train my own checkpoint in a reasonable time so I always use other peoples.

I have merged two checkpoints before if I want elements from both, that is a fast operation if you average the weights.

I'd recommend up-scaling with the SD upscale add-on using the same seed 10 iterations, .2 - .35 on strength
 
Jul 26, 2022
32
17
Civitai probably has the most, I train my own but I have used a few other people have made also...I dont have the power needed to train my own checkpoint in a reasonable time so I always use other peoples.

I have merged two checkpoints before if I want elements from both, that is a fast operation if you average the weights
Same I Too except I don't merge them but my loras doesn't look as good as Yours
 
  • Like
Reactions: felldude

felldude

Member
Aug 26, 2017
448
1,399
Same I Too except I don't merge them but my loras doesn't look as good as Yours
For every published lora I have, I have 4 that are not that great

Taking the time to manually focus and crop the images, as well as edit the deepboro output helps

I use realvision or uberporn merge of realvison for photo real
Dreamshaper sometimes for painting or fantasy
Orangemix for Anime
 
  • Like
Reactions: Hot-HighSchool-GaL