depends upon the art you're trying to do. i find i get far better details with a latent upscaler, but also tend to warp images far harder. its a double edged sword. if you do any anime featuring things such as cybernetic/cyberpunk or retro style with insanely fine details, latent is ultimately the better way to go.
Very nice tip ! didn't think of that. I usually go with freeU (the S1 and S2 parameters) to boost the sharpness and the "RescaleCFG" node to boost up the model by a multiplier of 0.5. But your way of doing is genius ! I will have to try this in my comfyui workflow :)
Do you know why that is? I've also been testing and they're slightly different. I'm not getting the same blurred or sharp images as you, but there's definitely some difference but I don't know if I should also just stay in the latent space as long as possible. Maybe it doesn't matter?
depends upon the art you're trying to do. i find i get far better details with a latent upscaler, but also tend to warp images far harder. its a double edged sword. if you do any anime featuring things such as cybernetic/cyberpunk or retro style with insanely fine details, latent is ultimately the better way to go.
Very nice tip ! didn't think of that. I usually go with freeU (the S1 and S2 parameters) to boost the sharpness and the "RescaleCFG" node to boost up the model by a multiplier of 0.5.
But your way of doing is genius ! I will have to try this in my comfyui workflow :)
awesome work
Thank you 😊
Do you know why that is? I've also been testing and they're slightly different. I'm not getting the same blurred or sharp images as you, but there's definitely some difference but I don't know if I should also just stay in the latent space as long as possible. Maybe it doesn't matter?
hey I tried it, it works very well ❤. But I prefer the model 4x_fatal_anime_500000_G
I tried to look for that to test out but haven’t seen it around.