Model based upscale scale #111
Replies: 1 comment 7 replies
-
|
These upscale models always upscale at a fixed ratio. You need to use the ImageScale node after if you want to downscale the image to something smaller. |
Beta Was this translation helpful? Give feedback.
-
|
Huh, so Comfy is just heavier on VRAM usage, alright. You are able to use scales in many other software though. I'm sure it can be done. |
Beta Was this translation helpful? Give feedback.
-
|
It's probably running out of memory in the VAEEncode node, if you switch it the VAEEncodeTiled the problem will likely go away. |
Beta Was this translation helpful? Give feedback.
-
Was wondering what that was. Lol |
Beta Was this translation helpful? Give feedback.
-
|
Loving VAEEncodeTiled & VAEDecodeTiled they are allowing me to get more size with my low vram. |
Beta Was this translation helpful? Give feedback.
-
Yeah the latest pack too @comfyanonymous did is really nice, more perfect then any I've seen in any other system. Like Even if you levels the image in photoshop to reveal the really low ends of the image, the tiling is super tight and faded nicely. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
There is no scale for the upscaling node for model based upscaling. This means most the good models will be forcing a user to do 4x when they may not even be able to do it under their GPU (especially if using it as part of HR-Fix like I do. Occasionally I even OOM on 4090 (weirdly by just 2mb lol).
Could we get a scalar input for the upscale node for a factor? Seems these models can handle arbitrary scales too like
1.5so it would be the best.Beta Was this translation helpful? Give feedback.
All reactions