Here is a some information, the idea of generating this is based on this other post #16986 I liked the idea but novel AI currently looks very obsolete.
first I generated some images of Saki and Yachie being alone, I used the loras of Saki and Yachie created by Imagine Breaker
then I used IMG2IMG + DPM++ 2S a Karras sampling method to blend the background into a single one. this was the most difficult part because some things the AI redrew better, but other parts were worse, it happened a lot with the face or the skirts. after generating 63 different images, I chose this last one, ironically 63 is my lucky number. https://cdn.discordapp.com/attachments/1042523770360967238/1109310593355227168/00065-SomethingV2_2_1326128075.png
then I used inpaiting to add Yachie's turtle shell and dragon tail, and improve Saki's wings and horse tail, this is the final result alternatively I made a different version that includes a hand drawn Lightning glare by me.
and before you say anything else, I improved the fingers on both versions, it's not perfect, but it's better than broken fingers.
Probably thanks to the combination of loras, without them it looks like this: asset #156407. Foxgirls for the fox god.
I see. That one looks cute too, though I definitely prefer the style of the upload for sure. It's amazing to see how far all of this stuff has come over the last several months.
I don't know why, but it gives me the feeling of having a "colored by markers" style so that's really cool. Also... fox girl, so that's always a huge plus for me.
Probably thanks to the combination of loras, without them it looks like this: asset #156407. Foxgirls for the fox god.
I don't know why, but it gives me the feeling of having a "colored by markers" style so that's really cool. Also... fox girl, so that's always a huge plus for me.
Took forever but I found the proper source. This is the source for the image (3rd image in the set) and the upload here was a very inferior version of it.
I have added an artist tag, it this marshmallow233, you can change it later. please, if you upload something generated by you add that tag + self_upload tag
This author's Pixiv account no longer exists. I just thought I'd post this one to save it from oblivion, since I like it and it doesn't seem to exist anymore. (His Twitter account still exists, but with that dlsite thumbnail at the top of his feed, I'm a bit surprised by that...)
No metadata on this, but I'm guessing it's NAI since a few of his other posts that I have archived have NovelAI metadata.
@tremolo_measure Yup. I used Anything 4.5 model. Metadata might not match 100% cause of inpainting
I have added an artist tag, it this marshmallow233, you can change it later. please, if you upload something generated by you add that tag + self_upload tag
Tag your the images you have generated with self_upload & make an artist tag for yourself (for example gamaz). You can easily change a tag to an artist tag by writing artist:gamaz to the tag box.
I guess shilling for my own model was a success, huh?
Certainly was, I've been dabbling in a few models, this one seems to run pretty well, although it's biased towards a couple things that I wish were more variable (In particular, lighting), it overall has such quality that I'm more than happy with it.
@ANJU Wasn't really my intention here, but it does look like early pregnancy I guess. I just wanted a bit of belly fat (pregnant was still in the wildcard, but wasn't used here).
Also this is my last upload for something like this. I'm going to try other things with this character, as I'm making a better LoRA for this character. In addition this was also an experiment of motion, so she is punching and bringing her arm into her side to chamber for another move like in a martial arts match.
I forgot to put logos and text in negatives. Then this turned out so good that it took me a while to notice after uploading that there are logos all over the bottom. Would delete this is that was an option.
Hey, uploader here. I made a mistake. I mean the Maxwell architecture, not the Pascal architecture. My bad! It has something to do with Cuda 5.x being converted differently, leading to a different starting sampling picture. Anything _starting_ from the Pascel architecture mostly results in the same thing, and I can't wait to try it out when I get my hands on it!
Absolutely no way, this looks amazing! You took my prompt and just made it absolutely beautiful, I was really struggling with it, at some point it just started degrading and I couldn't find the root of it. Love it, great model, good tweaks, and most of all, nice lack of double ears.
Thanks! I saw your commentary about an outdated berrymix and wondered what it would look like with a different model. I really think you should consider upgrading models- a lot of the newer ones are really quite better. Also with negative embeddings you don't really have to go with the giant negative prompt anymore. You can generally just throw one of them in there. The one I used was https://huggingface.co/datasets/gsdf/EasyNegative, but there are a lot.
Absolutely no way, this looks amazing! You took my prompt and just made it absolutely beautiful, I was really struggling with it, at some point it just started degrading and I couldn't find the root of it. Love it, great model, good tweaks, and most of all, nice lack of double ears.
Dayum, really lovely, especially how almost looks like the anime but with a bit more detail around the eyes and hair, I guess it was trained by anime screenshots since there is not too much material about Sunomiya.
Hello, what is your model? I have been searching for a long time but have not found a model named lora:artist-fishine
not sure if i can post a link here, but this specific fishine lora can be found in the gayshit link in the top post of the /hcg/ thread on /h/, ctrl+f fishine in the link and you will find it, i believe it is the first one. there are two other fishine loras, one named artist-hirame, and another named Hirame-v5. those are not the ones i use, but they still work okay. clarity_2 VAE is also required to replicate this output (can be found on huggingface schisim/clarity), but it doesn't show up in the metadata. there is also no other post-processing done here besides basic highres fix.
Amazing image and an amazing mix! I was about to ask if it wasn't called sweet soup before, because that's how I tagged it before but I see you already retagged my post, thanks.
Wow, a really old Favorite of mine. Didnt expected to see an AI version of Irina here. Really underrated Waifu. I just think her breasts are a bit more bigger in the anime but great image.
Having looked into it, the ambiguity here seems to be intentional. The "Average AI enjoyer" used to appear on their twitter but has since been edited out. Further, they tried uploading some stuff over on danbo and, when confronted about it, stated (twice) "I'm not saying that I generate pictures or draw them, I let people think the way they like." Imo this seems like a kinda scuzzy way to go about things so I'm probably not gonna upload any more from them.
I'm not surprised. There's quite a number of people making AI images that try to pass it off as their own original works, which is very annoying and sometimes only makes it harder to differentiate.
The money thing is a matter I don't like, but it wouldn't be as bad if they would just be honest about it everywhere instead of acting like how they did in their danbo comments, trying to (badly) muddy things like that.
I see. I looked around the artist's sites, and I found no indication of their work being AI. Plus, I couldn't find any typical AI mistakes, but I suppose my eyes aren't as attuned as others' might be...
In addition to the "I will share with you my own private Lora's that I use including my style" on their patreon and the "Average AI enjoyer" on their deviantart, I found two images with small errors. In this image you can see the sidelock on the left merging into the shadow and in this image the nails are weird. It's worth keeping in mind that as image models advance, we should expect the kinds of mistakes models are making today to become less frequent.
Having looked into it, the ambiguity here seems to be intentional. The "Average AI enjoyer" used to appear on their twitter but has since been edited out. Further, they tried uploading some stuff over on danbo and, when confronted about it, stated (twice) "I'm not saying that I generate pictures or draw them, I let people think the way they like." Imo this seems like a kinda scuzzy way to go about things so I'm probably not gonna upload any more from them.
I see. I looked around the artist's sites, and I found no indication of their work being AI. Plus, I couldn't find any typical AI mistakes, but I suppose my eyes aren't as attuned as others' might be...
Admittedly I'm not great at differentiating a fair amount either, but I haven't seen any proof to show me their stuff isn't AI generated. Their patreon has language that suggest they do AI stuff mainly and their deviantart profile mentions being into AI stuff too and many of their uploads there are 1024x1024 so it seems to me another sign to be that it's AI stuff.
They probably are just very attentive for any errors or something. If there's proof that confirms otherwise I won't have a problem deleting this.
It is. On their patreon the wording is clear that they use specific loras and a lot of their posts on Twitter have a similar style to this one.
I see. I looked around the artist's sites, and I found no indication of their work being AI. Plus, I couldn't find any typical AI mistakes, but I suppose my eyes aren't as attuned as others' might be...
As newer models come out and improve, I have little doubt that AI backgrounds will also improve too, however I would say expecting every detail to be correct every time isn't reasonable.. at this time anyway.
The background in this image doesn't look that bad to me, even if it may not be 100% correct.
Personally I think backgrounds deserve leeway. Asking for a 100% perfect background with AI is something of an impossibility to say the least.
I totally agree, especially with AI backgrounds! Though I also think that backgrounds should make sense, which is why I was concerned for a sec about the background possibly breaking the rules of eyesight.
I'm not sure if I understand. Are you saying that from the horizontal perspective right to left, it looks sloped due to the left trees looking beneath the right trees? If yes, I can also reason that the ground, up until the river, is sloping down (and therefore, this picture makes sense perspective wise).
Well, I mean, at least that's how I saw this image when I generated it.
Speaking of flagged status, the trees on the left side of the picture are tilted to the right, so the ground is sloped down, and I think there's enough distance to see the lake.
I'm not sure if I understand. Are you saying that from the horizontal perspective right to left, it looks sloped due to the left trees looking beneath the right trees? If yes, I can also reason that the ground, up until the river, is sloping down (and therefore, this picture makes sense perspective wise).
Speaking of flagged status, the trees on the left side of the picture are tilted to the right, so the ground is sloped down, and I think there's enough distance to see the lake.
Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based Based.
"You know, I know this steak doesn’t exist. I know that when I put it in my mouth, the Matrix is telling my brain that it is juicy and delicious. After nine years, you know what I realize? Ignorance is bliss."
This comment also said img2img. I'm new here and generated some images with AbyssOrangeMix. But AIbooru has been really motivating my research into stable diffusion 😸
