Looking through their Twitter, I wonder if they're still using Stable Diffusion to generate these. They have tweets from Early October of generations from it (and Waifu Diffusion.)
Continuing on the quest to generate ever smaller girls, I arrive at... *squints*... what appears to be someone embedded in a water droplet on top of some sort of rugged surface; a leaf, maybe?
As a slight note; The prompt will very likely not generate the exact same image. I slightly edited another version and this was a variation from that edited version.
Adding onto @x_u's comment, if you're using novelai, then try using 1 boy and use the curly brackets "{ }" on it. They add more strength to whatever tag(s), sentences, or descriptors they're used on. It's especially good for things that have trouble generating so you can "force" it to hopefully get the desired result.
Sorry, this image was generated using not only prompt, but also "img2img". First I uploaded an image such as "1 girl image" generated by myself or favoirite images found in Danbooru. Secondary I modified it using 'Edit Image'. For example, I added a shadow looking like a man's silhouette. And then I generated images with Strength 5~7, repeated several times to complete it. As you said, "1boy" frequently disappears... but if you use 'Edit Image', he might show up.
Hello, Beautiful picture but i can't seem too 100% replicate the model, could you make a more detailed recipe so i know exactly what interpolation method you used. Thank you
This is how (I think) I made it. My ui crashed like three times during this, so I included hashes if you want to try and debug it. All steps with 16bit floats. 1. weighted sum - A: anything-v3.0-pruned-fp16 [38c1ebe3] - B: sd-1.5-pruned-emaonly [81761151] - M: 0.05 2. add difference - A: mix_from_step_1 [21aa6034] - B: f222 [44bf0551] - C: sd-1.5-pruned-emaonly - M: 1 3. weighted sum - A: mix_from_step_2 [21aa6034] (for whatever reason it has the same hash as before, but the results are different) - B: trinart2_step115000 [f1c7e952] - M: 0.25
Hello, Beautiful picture but i can't seem too 100% replicate the model, could you make a more detailed recipe so i know exactly what interpolation method you used. Thank you
To explain step 6 in more detail: The part where you type 0.1-1.0 (+0.1) is a fancy way of telling webui to create a series of images, starting with one at 0.1 denoising (almost identical to the masked section, paying very little respect to the new text prompt) all the way to 1.0 (respecting only the new text prompt and ignoring the masked section entirely), in +0.1 increments (0.1, 0.2, 0.3, etc., up to 1.0).
You could change that +0.1 to +0.05 and get twice as many images, or to +0.01 to get 10 times as many. It's up to an agreement between you and your graphics card.
Also, this is a BE example because that's the most obvious thing to do with this sequence, but it's pretty obvious how creative use of prompts can modify other things instead. I'll leave that up to your imagination.
I updated your character tag to tetra_(tetra_ai). The recommended case is to have the owner's name (or nickname) as the qualifier instead of the generic "OC."
Hey, I really enjoy your work and you're my big inspiration frankly speaking. I was trying to recreate your style to find a really good starting point, especially when it comes to Model Merging, as it affects work greatly from my newbie understanding. I don't think that I'm doing something wrong as I checked thoroughly every step(and yes I was following all prompts, merged models and so on) and my result is different quite a lot. I'm not talking solely about this one but also about cute antler girls in more realistic style and many others :D. Is there a chance that you would think about uploading your most used Model Checkpoints(I noticed just one shared in other comment's section) to some site such as megaupload? Or would you rather not share with it completely? Of course if it isn't bothersome to ask such thing..
Best regards from your little fan Nishi!~
Hard to say what exactly is the problem. I mess around with versions, merges and setting a lot and there are and were some bugs that changed how images were generated. Here's a list of what I think might be going on: 1. I messed up instructions for one of my merges (one that is on mega). I have no idea how to recreate it now, sorry. This could have happened more than once so it might be my fault. 2. Some of my merges (like the one in this upload) were made using old ui. Try swapping models in the merge to: `trinart2_step115000_0.2 + nai_0.8` 3. You probably did this already, but check the settings twice. If I don't post something in the prompt I probably used CLIP: 1, firstpass: 0x0, hypernetwork: none.
If all else fails, you (or anyone else really) are free to message me with a particular image you want to recreate with a link to catbox or something with your result and I'll try to help.
Hey, I really enjoy your work and you're my big inspiration frankly speaking. I was trying to recreate your style to find a really good starting point, especially when it comes to Model Merging, as it affects work greatly from my newbie understanding. I don't think that I'm doing something wrong as I checked thoroughly every step(and yes I was following all prompts, merged models and so on) and my result is different quite a lot. I'm not talking solely about this one but also about cute antler girls in more realistic style and many others :D. Is there a chance that you would think about uploading your most used Model Checkpoints(I noticed just one shared in other comment's section) to some site such as megaupload? Or would you rather not share with it completely? Of course if it isn't bothersome to ask such thing..
That makes sense, I was thinking about votes in a different way. Probably because all the posts with negative votes have been deleted pretty quickly, (usually because they actually were breaking the rules). So i guess i just made a mental connection without questioning it much.
In any case, does something like this qualify as guro in your mind? Personally don't really want to see it if I don't have to, though I can appreciate that others might like it. Monster Girl is a bit broad because anything with even a slight amount of "monster" to it applies, like a slime girl or harpy etc so you can't really set up filtering based on that.
I usually only think of guro as extreme violence, but I know others (including pixiv) have differing definitions.
People are downvoting it presumably because they don't like it. Score is a metric of popularity, not whether or not it breaks the rules. Whether or not it ends up approved comes down to if an approver likes it enough, as it doesn't seem to break the rules.
That makes sense, I was thinking about votes in a different way. Probably because all the posts with negative votes have been deleted pretty quickly, (usually because they actually were breaking the rules). So i guess i just made a mental connection without questioning it much.
In any case, does something like this qualify as guro in your mind? Personally don't really want to see it if I don't have to, though I can appreciate that others might like it. Monster Girl is a bit broad because anything with even a slight amount of "monster" to it applies, like a slime girl or harpy etc so you can't really set up filtering based on that.
Might not be to my tastes but i dont understand why people are downvoting as it IS ai art, however niche and not "nice" it may be to some. It is a little disturbing to some so i'd probably label future posts like this with the guro tag (i believe it classifies) so it can be filtered by default for those who don't want to see that.
Dont let it discourage you though.
Edit: It looks like the wiki description for what qualifies as guro is different on here than on Gelbooru where its more broad. Gelbooru excerpt:
Would love a staff take on this as i feel like it does belong but might not be welcomed by everybody
People are downvoting it presumably because they don't like it. Score is a metric of popularity, not whether or not it breaks the rules. Whether or not it ends up approved comes down to if an approver likes it enough, as it doesn't seem to break the rules.
Might not be to my tastes but i dont understand why people are downvoting as it IS ai art, however niche and not "nice" it may be to some. It is a little disturbing to some so i'd probably label future posts like this with the guro tag (i believe it classifies) so it can be filtered by default for those who don't want to see that.
Dont let it discourage you though.
Edit: It looks like the wiki description for what qualifies as guro is different on here than on Gelbooru where its more broad. Gelbooru excerpt:
Extreme graphic violence is part of it but things like horrifying monsters and body horror are also guro. Basically anything that is simultaneously erotic and grotesque qualifies as guro."
Model is: Save as float16 but only at the last step (for same results, float32 gives sometimes dissimilar results) [Add Difference] Stable Diffusion 1.5 + (Anything 3.0 - Stable Diffusion 1.4) [M=1] -> mergeStep1 [Add Difference] mergeStep1 + (Zeipher_F222 - Stable Diffusion 1.5) [M=1] -> mergeStep2 [Weighted Sum] mergeStep2 + TrinArt2 [M=0.25] -> Final Model
Very neat. I'll have to try a bunch of these on various models. The stuff getting generated on Yiffy-e18 is often very samey since it seems like half the prompts anyone uses invoke Pino Daeni, and I'm guilty of relying on him too much myself in that model.
I went through and made my picks from the different tabs on that. Of course, all the examples included with that are generated from just SD1.5, and doesnt really show fully how it would impact a more specialized model (such as NovelAI). That's why i made this.
Thanks. All this stuff is so new. Learning the basics is a lot, and now there's extensions on top of it... Overwhelming in a fun way.
I went through and made my picks from the different tabs on that. Of course, all the examples included with that are generated from just SD1.5, and doesnt really show fully how it would impact a more specialized model (such as NovelAI). That's why i made this.
I could stuff any number of general tags in the tag list but it seems pointless given how many images there are. Also, looks like i popped the "Incredibly_Absurdres" cherry!
Maybe I did mess up somewhere, since you're not the first person getting a different hash. If you want, some anons in the thread wanted it, so it's on mega here: `file/nZtz0LZL#ExSHp7icsZedxOH_yRUOKAliPGfKRsWiOYHqULZy9Yo`
Legend. Thanks :).
Iduno if saving it as float16 changes the hash, but probably? So that may be what it is. I Assume thats why its only 2GB vs most of my other models which are 3.9
The model you have is really good, i tried to make it for myself using the instructions in the model name but i don't get the same hash, it does produce similar results though. Was wondering if you can give some more info on it 🤔
Maybe I did mess up somewhere, since you're not the first person getting a different hash. If you want, some anons in the thread wanted it, so it's on mega here: `file/nZtz0LZL#ExSHp7icsZedxOH_yRUOKAliPGfKRsWiOYHqULZy9Yo`
