Cant repeat the result ,which i generated on civita ai.Cant get how to use loras from the additional resources tab on my local version of stable diffusion.

[Aflexg]

Hello.I cant repeat the result (on my local version of the sd 1.5), which i generated on civita ai even with the same model, promt and seed.

It seems the problem is associated with the additional resources tab. There is 2 loras in this tab and they are not included in the promt. So, in my point of view, i always copy incorrect/incomplete promt.

Besides this there is a trigger word: pencil sketh, which is not included in the promt too.

To solve this i tried to add these loras (and a keyword) in my promt, but , unfortunatelly, it didnt help.

Could you explain me, please, how to use loras from the additional resources tab and a keyword on the local version of sd correctly to repeat exactly the same result,which was generated on civita ai ?

Results of the generation:

Result form the civita ai:
a girl
easynegative, (bad-hands-5:1.5), (loli:1.2), (worst quality, low quality:1.4), blurry
My results:

1. a girl
   easynegative, (bad-hands-5:1.5), (loli:1.2), (worst quality, low quality:1.4), blurry

2. a girl, lora:Pencil_Sketch:1, lora:sketchkha:1
   easynegative, (bad-hands-5:1.5), (loli:1.2), (worst quality, low quality:1.4), blurry

(Commentary: it looks almost the same, but this is not enough, because it works only with this extremly simple promt.Complex promt leads to the big disparities between the civita ai result and my result)

3. a girl ,pencil sketch ,lora:Pencil_Sketch:1, lora:sketchkha:1
   easynegative, (bad-hands-5:1.5), (loli:1.2), (worst quality, low quality:1.4), blurry

[w-e-w]

we do not know how civita ai generate their images
likely they are using a different implementation from webui
thet or you're missing some hidden default settings

[dilectiogames]

there is a chance that civitai may insert hidden tags to prevent generating illegal content.

apart of that look into clipskip and env options.

also try setting the questionable tag at 0.0 instead 1.2 but don't remove it. see if you get a closer resemblance with the civitai output.

and if you or them or both of you have xformers on then is pointless, you will never get the same exact output.

[w-e-w]

and if you or them or both of you have xformers on then is pointless, you will never get the same exact output.

no, xformers should be deterministic as of v0.20.x

[Aflexg]

Thank you for your replies.I could find the solution .In my case, Civita ai doesnt recognise easynegative embeding, that's why the real negative promt of the picture from civta ai is: (bad-hands-5:1.5), (loli:1.2), (worst quality, low quality:1.4), blurry.

So promts are a bit diffrent and ,as a result, pictures are a bit different too.