Stable Diffusion Generation Dump

[Jemnite]

It's mega cute.

Jemnite/WagaShikashi · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

Yatsuhashi (八橋Mix) - v1.0 | Stable Diffusion Checkpoint | Civitai

A more refined MBW-COS Aurora_N dpep5mzpikaz+tmnd This mix is intended to make cute and interesting dynamic images. It has a very cute and thick te...

civitai.com

Basically it's like a less onerous to use version of MBW-COS Aurora_N dpep5mzpikaz+tmnd. There's a few differences, MBW-COS Aurora_N dpep5mzpikaz+tmnd is sharper, got less pallet swatche like shading, more variant and dynamic, but given the fact that you have to load like 5 loras every picture and this causes a massive slowdown in compute time, I've elcted to just call this one the replacement.

 

[Jemnite]

I live to image dump lmao.

As you might have suspected (or know, if you follow my civitai which actually uploaded first for once), new GyozaMix ver.

Jemnite/GyoZanMix · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

Store Bought Gyoza (餃子Mix) - WinterHoliday2023Edition | Stable Diffusion Checkpoint | Civitai

饺子皮是工场做的可是我答应质量很好。比手工还好吃！ Description: A stylized anime model. This model is the result of attempts to get the textures from MouseyMix , a model th...

civitai.com

These aren't all of them, just the ones that I think are the best and also worksafe.

I wrote a long screed on this somewhere where you guys will never see it, but to put it short it turned into a 2.5D model thanks to all the lighting upgrades and for once, even though I'm not a huge fan of 2.5D models, I'm okay with it. The important thing I've come to realize about 2.5D models is that the lighting work (specifically the dynamicism potential) needs to be able to keep up with the texture complexity. AOM3 sucks because even though the textures are halfway into 3D, the lighting is pretty shallow and uniform, and that makes it seem plasticky. You need dark shadows and superbright glare to make it actually feel like it's 2.5D without being uncanny.

 

[Jemnite]

Click to shrink...

Experimenting to see if I can pump out one last Gyozamix before SDXL comes out and everyone moves off 1.5. Most of the things I used it to generate aren't SFW but here's a shot by shot comparison.


Managed to peel it back to a 2D style somewhat rather than a 2.5D style. IMO the eyes are generally better across a broad spectrum, and have more of that V3.0 DNA in them.

 

[waito_x]

Very interesting, Jemnite. Please keep at it I look forward to seeing more.

 

[Jemnite]

Very interesting, Jemnite. Please keep at it I look forward to seeing more.

Click to shrink...

I honestly feel like it's still too bright at this point. I'll try injecting some noise offset to try to peel down the brightness levels a bit. Part of it is due to the huge amount of lighting loras I crammed in there. When I was playing around with the Lunar Radiance model dropping in noise offset seemed to move the model back towards a darker tint and reshaped the lighting contrast rather than just overall brightness so it might be worth a shot.

 

[Jemnite]

I honestly feel like it's still too bright at this point. I'll try injecting some noise offset to try to peel down the brightness levels a bit. Part of it is due to the huge amount of lighting loras I crammed in there. When I was playing around with the Lunar Radiance model dropping in noise offset seemed to move the model back towards a darker tint and reshaped the lighting contrast rather than just overall brightness so it might be worth a shot.

Click to shrink...

No wait, I'm an idiot I just loaded the wrong VAE.

Lmao yeah ofc they're going to look super bright if I unload the baked VAE for a saturation VAE.

 

[Kingcrusader]

No wait, I'm an idiot I just loaded the wrong VAE.

Lmao yeah ofc they're going to look super bright if I unload the baked VAE for a saturation VAE.

Click to shrink...

Huh, it actually looks better when it is a little dark. You also kinda changed the girl's look a little bit.

 

[Jemnite]

A refresh on an old classic anime.



You can download the V5 if you want, it's in the HuggingSpace repo actually properly named for once. I'll probably drop it on Civitai tomorrow after experimenting with it some more.

 

[Jemnite]

So, I uploaded V5.

Store Bought Gyoza (餃子Mix) - VV | Stable Diffusion Checkpoint | Civitai

饺子皮是工场做的可是我答应质量很好。比手工还好吃！ Description: A stylized anime model. This model is the result of attempts to get the textures from MouseyMix , a model th...

civitai.com

I have more example images but uh, they're not exactly sfw. I put them up on pixiv because they're a buncha degenerates so if you're really interested you can go look them up there. Someone apparently took one of the generations and used it to start a controversy on 4chan's /ic/ board (Apocal told me about it lmao) which I don't know if it's a good thing or a bad thing lmao.

Anyway have some spooky boogie spyders

 

[Exponent]

I put them up on pixiv because they're a buncha degenerates so if you're really interested you can go look them up there. Someone apparently took one of the generations and used it to start a controversy on 4chan's /ic/ board (Apocal told me about it lmao) which I don't know if it's a good thing or a bad thing lmao.

Click to shrink...

Do you have a wider gallery of your SFW art on Pixiv or anywhere else?

 

[Jemnite]

Lua's Night. It behaves very well at points, and somewhat poorly at other points. I think if you can bring the lighting strongpoints to leverage it's a fantastic model but otherwise it's somewhat not super responsive to prompts and goes off the rails a lot to do its own thing.

Lunar Radiance (月光Mix) - Lua's Night | Stable Diffusion Checkpoint | Civitai

Dream... not of what you are, but of what you want to be. A mix targeted at enhancing lighting effects. The base checkpoint is also CLIP expanded u...

civitai.com

 

[Jemnite]

ImpastoMix. No, the name isn't a reference about being sus. It's actually from the term Impasto, which is a oilpainting technique where you use a brush or knife to lay paint in extremely thickly. As the name might suggest it's excellent at mimicking the texture of oil paintings and is whacked out in terms of how well it does certain feels and styles.

Take a look.

ImpastoMix - v2.0 | Stable Diffusion Checkpoint | Civitai

A mix that mimics impasto-style oilpainting. Texture heavy but style neutral, most style loras are translated as if the artist was an oilpainter. G...

civitai.com

Jemnite/FauxMediums · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

Spoiler: Bonus

 

[Jemnite]

BTW do you guys actually want to see slightly risque gens? I've been avoiding posting anything that shows more than leg, midriff or cleavage, but the oilpaint model has showsn itself surprisingly adept at generating musclegirls so I've just ended up with a fuckton of scantily clad buff women in swimsuits/sportswear.

 

[Kingcrusader]

BTW do you guys actually want to see slightly risque gens? I've been avoiding posting anything that shows more than leg, midriff or cleavage, but the oilpaint model has showsn itself surprisingly adept at generating musclegirls so I've just ended up with a fuckton of scantily clad buff women in swimsuits/sportswear.

Click to shrink...

I mean, not really. Though, I would like to see more bricc girls with muscles on them. Heck, add monster girls if you can.

 

[Exponent]

Some space stuff?

 

[Jemnite]

Some space stuff?

Click to shrink...

I don't take requests. But I do have some old shit lying around that you can have if you want.

 

[waito_x]

I mean, not really. Though, I would like to see more bricc girls with muscles on them. Heck, add monster girls if you can.

Click to shrink...

I am a big fan of the monster girl So, I noticed the past several months have gone from image artifacts to really clean.

for example

While the improvement is really beneficialt to the craft, this does mean that there is less to confuse / distract / deter the diffusion-GAN procedures from throwing in added detail regarding prompts that accidentally or intentionally call upon various levels of undress.

 

[Kingcrusader]

I am a big fan of the monster girl So, I noticed the past several months have gone from image artifacts to really clean.

for example

While the improvement is really beneficialt to the craft, this does mean that there is less to confuse / distract / deter the diffusion-GAN procedures from throwing in added detail regarding prompts that accidentally or intentionally call upon various levels of undress.

Click to shrink...

??? I can't tell what you mean by this. Don't you mean it's possible to do a type of undress?

 

[waito_x]

Seemingly innocuous text prompt >>

Diffusion model heuristics >>

Less clutter, more skin. Resulting image NSFW.

 

[Kingcrusader]

Seemingly innocuous text prompt >>

Diffusion model heuristics >>

Less clutter, more skin. Resulting image NSFW.

Click to shrink...

Oh ok. Thanks. This makes it more clear.

 

[waito_x]

Oh ok. Thanks. This makes it more clear.

Click to shrink...

Yep, what the AI usually throws out because too much of the allocated cycles are spent with cluttering the foreground (specifically) versus focusing like a professional photographer.

edit: Jemnite, if the look of the cited picture was intended and done through in-painting, know then I mistook that for image artifacts.

 

[Exponent]

I don't take requests. But I do have some old shit lying around that you can have if you want.

Click to shrink...

I didn't notice the attach pictures somehow, so pardon my utter silence. Thanks for uploading.

 

[steamrick]

edit: Jemnite, if the look of the cited picture was intended and done through in-painting, know then I mistook that for image artifacts.

Click to shrink...

If you look at the prompt of the image you used, the look of the picture was intended and actively prompted for... in addition to the fact that Jemnite tends towards checkpoints that produce busy pictures in the first place.

Here's the prompts with the parts liable to produce a busy looking picture highlighted...

masterpiece,
best quality, 1girl, (colorful),(delicate eyes and face), volumatic
light, ray tracing, bust shot ,extremely detailed CG unity 8k
wallpaper,solo,smile,intricate skirt,((flying petal)),(Flowery meadow)
sky, cloudy_sky, moonlight, moon, night, (dark theme:1.3), light,
fantasy, windy, magic sparks, dark castle,white hair

Click to shrink...

If you want 'really clean', there's anime models that'll do more flat textures. ToonYou comes to mind.

 

[waito_x]

Things like +solo +highres --no bad anatomy usually do nicely.

 

[Kingcrusader]

Yep, what the AI usually throws out because too much of the allocated cycles are spent with cluttering the foreground (specifically) versus focusing like a professional photographer.

Click to shrink...

I can see that, because the girl in the image looks to be standing on both her feet, while everything else is tilted. But i guess that's how the AI operates, because I don't know how it works when it comes to stuff like this.