Great approach! I would mask off some details on the skin or maybe just blend color instead of luminance on an external editor, keeping the whole detail on clothes, lips, irises..., but overall it looks pretty good!
Agree with others. This is one of the best attempts at this upscaling and enhancing. I will say the final result is a bit blurry… and doesn’t get some textures just right. But it’s VERY close to magnific. I think some slight tweaking at the end and maybe re shrink the image and dump it through a simple SDupscale again to sharpen details might solve that
Is an absolute great output, look to the big picture and see the amount of time this is saving producing such quality! I think you're looking for an almighty workflow that solve all at once, and that is not gonna happen because this tools works as an extension of the artist not as a replacement of them. So to get the result your looking for is as easy as to take the output image to Photoshop and give it a couple tweaks using Neural Filters. Some of them dedicated to photography.
this works well with sd images but not regular photos taken off the web. Maybe you could try adding an auto-tagger node to better describe images that do not have a prompt? Not sure what else I could do but the results were very disappointing. But on stable diffusion images they were great.
I'm getting this error: Error occurred when executing SamplerCustom: 'dict' object has no attribute 'cond_hint' For some reason the samplercustom node is stopping the whole process and I haven't yet found a solution
I've got a 2080Ti (11GB VRAM), and it doesn't cope with this, get cuda out of memory errors on the Custom Sampler. Should have a 3090 arriving tomorrow or Wednesday, so look forward to giving it a try once I've replaced the GPU.
That's a great upscaler. Keep the details and enhance at the same time. Great job sir and thanks for sharing🙂
Thanks Ed, Your team is doing a great job!
love this. y'all amazing and thanks for breaking it down
Great approach! I would mask off some details on the skin or maybe just blend color instead of luminance on an external editor, keeping the whole detail on clothes, lips, irises..., but overall it looks pretty good!
RTX 4060ti 16g --- 1hour.22min ))))
the workflow stops at "sampler Custom"
'NoneType' object has no attribute 'shape'
Agree with others. This is one of the best attempts at this upscaling and enhancing.
I will say the final result is a bit blurry… and doesn’t get some textures just right.
But it’s VERY close to magnific. I think some slight tweaking at the end and maybe re shrink the image and dump it through a simple SDupscale again to sharpen details might solve that
Is an absolute great output, look to the big picture and see the amount of time this is saving producing such quality!
I think you're looking for an almighty workflow that solve all at once, and that is not gonna happen because this tools works as an extension of the artist not as a replacement of them. So to get the result your looking for is as easy as to take the output image to Photoshop and give it a couple tweaks using Neural Filters. Some of them dedicated to photography.
this works well with sd images but not regular photos taken off the web. Maybe you could try adding an auto-tagger node to better describe images that do not have a prompt? Not sure what else I could do but the results were very disappointing. But on stable diffusion images they were great.
I'm getting this error: Error occurred when executing SamplerCustom:
'dict' object has no attribute 'cond_hint'
For some reason the samplercustom node is stopping the whole process and I haven't yet found a solution
I've got a 2080Ti (11GB VRAM), and it doesn't cope with this, get cuda out of memory errors on the Custom Sampler. Should have a 3090 arriving tomorrow or Wednesday, so look forward to giving it a try once I've replaced the GPU.
I can only wish 😢
Got the replacement card fitted and gave this another try, really like it :) I swapped everything out to SDXL though.
I unfortunately am getting an out of memory error.
Maybe start with a lower resolution image
@@AIFuzz59 That seems to be the case. Thank you for your feedback and time and effort.
@@AIFuzz59 do you have a blog with detailed description of the flow of information from input image to output image?