>>27592 Left side tags are added to the end of the negative prompt. Right side is normal negative. I'm not experiencing it at all, unless I'm blind. Metadata is in stealth PNGinfo.

>>27606 nta but i see it, most of the eyes look far apart enough to be noticeable

Tentacles ClowCards.

retard here, how do I add the cfg++ samplers? would be using them on noob if that helps.

>>27609 imo they're snakeoil but they either come with reforge on dev_upstream branch or you can install them on comfy

>>27594 Btw if any of you are generating tags with wd14 tagger extension, make sure min tag fraction is set to 0 instead of default 0.05, that setting removes tags featured in less than 5% of the images

>>27582 box for that last one?

angel and the chink alliance delivered where Jordach got thousands and has literally nothing to show for it

https://files.catbox.moe/it83o0.png test with female anatomy

>>27606 Could this be it? I have a few artist mixes with `gloss` in them, for example the first two X/Y plots of >>27592 is genned with this one: (quasarcake:0.7), (sweetonedollar, mery \(yangmalgage\):0.75), gloss, observerz, (s16xue, fizrotart:1.1) It should be mapped to the artist `gloss_(hotglossed)` on danbooru, but on e621 it's the name of a furry character https://danbooru.donmai.us/posts?tags=gloss_%28hotglossed%29 https://e621.net/posts?tags=gloss

>>27612 https://files.catbox.moe/tphup8.png

>>27617 I'm not sure if that would be enough to cause it, like 5 images that have the token sequence association vs 150+ for the artist would be strange. Noob 1.0 in general feels very sensitive to negatives, moreso than any other model. You can really see it in the X/Ys I posted, I think they were all 6 CFG and the furry negs exacerbate the color burn.

>>27619 It's 7 images with a lot of favorites, one of them have more than 400, their impact on the style will be boosted significantly by quality tags Even with uniform repeats my experience in lora training tells me that 7 out of 135 is enough to bake a feature into the style if not tagged properly >Noob 1.0 in general feels very sensitive to negatives I think furry negative is a cope to begin with: it consists of 3 random tags with high frequency on e621, not an official tag noob team added to the data, also negative prompts are at most as reliable as positive prompts, if you would expect the positively prompted sex posture to be nonsensical 20% of the time, why would you expect negative prompts to be reliable at the same or higher level And this is just v1.0 of eps-pred, the v-pred model will be trained on this with the e621 dataset intact

>>27615 boxes?

in other words: we're fucked lol they really just snatched defeat from the jaws of victory by adding the e621 scrape in

>>27626 >we're fucked not as if just a few months ago it was either paying 25 dollars for a locked down model with no loras or dealing with pony

>>27627 you know what i meant

>>27626 please calm down and take a break from the computer lest we end up like /h/ where people are acting like the ai porn generator not being ideal is the end of their lives

>>27629 very funny but you don't need to project

>>27630 *projects my dick into your ass*

>>27631 ahhh ahhh

>>27632 ahh ahh mistress*

Just use whatever you want Me? I don't care about it. Keep those finetune coming.

Noobai makes good Tanyas. Therefore good model.

>>27634 Sylviechads stay winning

>>27033 https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLg added abe tsukasa (noobvpredtest) added arako o (noobvpredtest) added shindol (noobvpredtest) added nyalia (noobvpredtest)

>>27639 catbox for that second one?