Open Omegadarling opened 2 years ago
Being a diffusion-based upscaler is not really relevant or adds any value compared to other upscalers. As far as I remember this is an unofficial code and has an unofficial model with so-so results. Have you tried it first?
If I'm gonna gank another upscaler and argue with @AUTOMATIC1111 about whether or not it's needed, I'd probably look at the successor to this paper by the same author, located here:
https://github.com/Janspiry/Palette-Image-to-Image-Diffusion-Models
But, after adding ScuNET upscaling, it'll be a hard sell unless there are clearly defined benefits.
Yeah, @d8ahazard, that image recovery in the Janspiry repository is phenomenal. And @victorca25, I'm just an artist who loves the hell out of these tools, not a coder. I'm at the mercy of what is made available by the wizards behind the curtain.
Lastly, ScuNET must have just been added, because I didn't notice it until after you mentioned it, @d8ahazard !
Yeah, @d8ahazard, that image recovery in the Janspiry repository is phenomenal. And @victorca25, I'm just an artist who loves the hell out of these tools, not a coder. I'm at the mercy of what is made available by the wizards behind the curtain.
Lastly, ScuNET must have just been added, because I didn't notice it until after you mentioned it, @d8ahazard !
Yep, we've now got ScuNET, BSRGAN, and LDSR. I find ScuNET is really good for denoising older pics as a starting point, and then using LDSR and GFPGAN create some great results. ;)
@d8ahazard I do a lot of work where I'm trying to add paint detail as I upscale. Most of the upscalers actually remove that kind of detail. I think that's why I've been hoping for something better.
UPDATE I just tried LDSR and it took a while, but it might be exactly what I'm looking for! It definitely added in a lot of brush strokes and detail.
Any chance you could implement this kind of upscaling technique? https://github.com/Janspiry/Image-Super-Resolution-via-Iterative-Refinement
It would use diffusion to upscale images generated by diffusion; what could be better?