Let people load checkpoints using load checkpoint node, do not force people to download and load stability sdxl stock weights

[1blackbar]

This is limiting and people who have lot of downloaded models have to download one more that wont be used by anything else but this node.
Just allow us to connect own checkpoint with vae and clip

LEt us also choose how many times it will upscale the image, there is no way to do that now.

[Kallamamran]

Hear hear!

I hate autodownloading models without letting me chose what model to download and where to put it. Sometimes I already have the model on disk and maybe I don't want to download it again!

Also... The node only upscales. I have no way of telling it waht to do. I just get a rendom upscale. I see limilted usefullness for this as it is now, but keep up the good work! Baby steps! :)

[rafstahelin]

second that would love to be able to use other models. some images great, some closeup portraits of faces get a little bit of hallucination away from linkeness, would b enice to be able to control that a bit more

[rafstahelin]

also, there is already a "diffusers" path in comfyui. why not place your diffusers there as well, so as to consolidate people's diffusers. just an idea

[yuvraj108c]

also, there is already a "diffusers" path in comfyui. why not place your diffusers there as well, so as to consolidate people's diffusers. just an idea

It already saves to models/diffusers path! so if the model already exists, it won't download again.

Loading .safetensors and other models should be possible in future updates (hopefully)

[yuvraj108c]

Also... The node only upscales. I have no way of telling it waht to do. I just get a rendom upscale. I see limilted usefullness for this as it is now, but keep up the good work! Baby steps! :)

Well, the original implementation of InvSR specialises only on upscaling/enhancing low res images.. what else do you expect from it?

[rafstahelin]

It's pretty cool all things considered. Seems to be worst on black and white images. And possibly it tends to smooth out skin and eyes when the image is complex or already has a high aesthetic value. Just speculating on certain outputs that were definitely below acceptable quality in areas of the image (ie eyes, skin)... though in a large number of outputs the results reminded me a little of Supir

[negative1ne]

Also... The node only upscales. I have no way of telling it waht to do. I just get a rendom upscale. I see limilted usefullness for this as it is now, but keep up the good work! Baby steps! :)

its not random, its a fixed 4x upscale, if you bothered to read what it actually does.

sounds like you're not sure what you are doing

[negative1ne]

second that would love to be able to use other models. some images great, some closeup portraits of faces get a little bit of hallucination away from linkeness, would b enice to be able to control that a bit more

just run another pass on it.
i use adetailer, which fixes it.

[anything like krea, and magnifique work also]

[KyleChan-bot]

Greetings! As I also experienved the same issue that I could not select the local difussers from the node, and I failed to run another loader to run pass it. I would appreciate a lot if you could kindly advise me how to do it.
Thank you in advance!

second that would love to be able to use other models. some images great, some closeup portraits of faces get a little bit of hallucination away from linkeness, would b enice to be able to control that a bit more

just run another pass on it. i use adetailer, which fixes it.

[anything like krea, and magnifique work also]

[1blackbar]

can you force it to do just 2x upscale ?

[yuvraj108c]

can you force it to do just 2x upscale ?

It can only do 4x: zsyOAOA/InvSR#9

[Xceciiiiiiilll]

I had to dl the models manually because my connection speed is slow. I put sd_turbo in models/diffusers but it's still stuck trying to download something when I run this workflow.