Open Kinglord opened 1 year ago
I think I'm an idiot nevermind.
Nevermind, not an idiot. I thought maybe I had it wired wrong using the model but then I downloaded your plot workflow and just did the same thing. Here you can see the difference between using the XY Input script vs using a normal lora loader, something is definitely going on here!
I edited my issue but after further testing, the outputs are the SAME as long as CLIP STR remains at 0. As soon as I use a clip value then the results start to deviate. Depending on the model/lora they can come close to converging at times, but I can't fine anything consistent about that.
Hey there,
I was extremely excited to try out the LORA Plot since I needed to work with a lot of LORAs, sadly as far as I can tell it won't work for me because the output with it is completely different from a normal Lora Loader node when CLIP STR is set above 0. I ran a bunch of tests trying to figure out if it was flipping values or doing something else crazy, but as far as I can tell it's something broken in the actual script process but I don't know where. To be clear, the output if I use the eff loader lora widgets matches the one from the XY Plot, but both are different from if I use an external lora loader when the LORA CLIP is above 0.
To make the test simple I used the same value for both my model and clip strength, but it still doesn't get anywhere close. I've attached sample images, one is using the lora at .5 with the normal lora loader, the other is using an XY input (not the plot, was trying to see if it's just the plot but it's all XY Lora inputs) and you can see the result is totally off. I tried a bunch of things to see if I could replicate the image coming out of the XY Grid Script but no luck at all. I'm really lost at this point. I'll attach my settings as well as the output so you can see the difference. I'm on the 10/18 version of Comfy, if maybe something recent broke it?