Stable Diffusion AI sneezes

[PollenFiend]

3 hours ago, hanahiru said:

I got a couple decent pics finally 

Nice!  Glad it's sorta kinda working out.

2 hours ago, MikiSan said:

whats that on the side o.o

lol...  I'm gonna go with... the AI imagining a little sneeze gnome/fairy doing its thing inside the girl's nose.

[MikiSan]

1 hour ago, PollenFiend said:

Nice!  Glad it's sorta kinda working out.

lol...  I'm gonna go with... the AI imagining a little sneeze gnome/fairy doing its thing inside the girl's nose.

XD ahhh ok 

[kushamizuki]

[MikiSan]

ooo still dont know what kind of prompt to use that doesnt give me a black box 

[webmeistro]

If you're running it locally and get a black screen every time, try running it with xformers off.

[MikiSan]

11 hours ago, webmeistro said:

If you're running it locally and get a black screen every time, try running it with xformers off.

how do i do that?

[webmeistro]

14 hours ago, MikiSan said:

how do i do that?

Try adding --no-half to the command line arguments.

[MikiSan]

3 hours ago, webmeistro said:

Try adding --no-half to the command line arguments.

oooh ok

[PollenFiend]

Hi!  I'm back with some new tips I've learned regarding inpainting.  Thought I'd share. 

After some more reading and experimenting, I've learned more about the "inpaint at full resolution" setting.  It turns out that even after upscaling a picture to something very high-res (e.g. 2160x2160), if your inpaint area is small, your computer will likely still be able to handle it as it only needs to work on the smaller inpainted area instead of the full-size image.  The benefit here is that the AI will first upscale the inpaint area to a much larger size before filling it in, and that will give it a lot more room to determine and add in small details.  This gives a better chance of drawing things accurately, like faces, if the inpaint area is relatively small compared to your total image size.

I'll post an example using my most recent pic (from this thread).  Here you see the inpaint mask area to redraw the face, in an image that is 800x800:

Then, here is inpainting with the "inpaint at full resolution" setting turned off, and the image resolution still set to 800x800.  As you can see, it isn't great.  The nose and mouth are kinda screwed up, and would need detailed work to get it better:

However, if the original image is upscaled, in this case to 2160x2160 and the inpaint area still just focuses on the face, if "inpaint at full resolution" is turned on, the AI has a lot more room to play with and determine what should look correct in terms of facial detail (even using the same prompts):

Note, the above pic still wouldn't be considered "done" but you might notice how the mouth and nose are generally more detailed and accurate.  So, using this technique I've experimented more (inpainting the face, or just the eyes, or just the nose, etc.) and made a series of images depicting Dawn's morning allergy attack.  Some elements are still hard to get consistent, particularly the color of the eyebrows and lips for some reason, but I'm generally happy with the end-result.  Full set of 8 variations will be on my Pixiv account.

I hope this helps your artwork.  Next, I'm now trying to learn about ControlNet which is something pretty new and exciting to help with posing human characters.

[PapaBear70]

2 hours ago, PollenFiend said:

Hi!  I'm back with some new tips I've learned regarding inpainting.  Thought I'd share. 

After some more reading and experimenting, I've learned more about the "inpaint at full resolution" setting.  It turns out that even after upscaling a picture to something very high-res (e.g. 2160x2160), if your inpaint area is small, your computer will likely still be able to handle it as it only needs to work on the smaller inpainted area instead of the full-size image.  The benefit here is that the AI will first upscale the inpaint area to a much larger size before filling it in, and that will give it a lot more room to determine and add in small details.  This gives a better chance of drawing things accurately, like faces, if the inpaint area is relatively small compared to your total image size.

I'll post an example using my most recent pic (from this thread).  Here you see the inpaint mask area to redraw the face, in an image that is 800x800:

Then, here is inpainting with the "inpaint at full resolution" setting turned off, and the image resolution still set to 800x800.  As you can see, it isn't great.  The nose and mouth are kinda screwed up, and would need detailed work to get it better:

However, if the original image is upscaled, in this case to 2160x2160 and the inpaint area still just focuses on the face, if "inpaint at full resolution" is turned on, the AI has a lot more room to play with and determine what should look correct in terms of facial detail (even using the same prompts):

Note, the above pic still wouldn't be considered "done" but you might notice how the mouth and nose are generally more detailed and accurate.  So, using this technique I've experimented more (inpainting the face, or just the eyes, or just the nose, etc.) and made a series of images depicting Dawn's morning allergy attack.  Some elements are still hard to get consistent, particularly the color of the eyebrows and lips for some reason, but I'm generally happy with the end-result.  Full set of 8 variations will be on my Pixiv account.

I hope this helps your artwork.  Next, I'm now trying to learn about ControlNet which is something pretty new and exciting to help with posing human characters.

Ohhh, I like this a lot. Her facial expression along with her pinkish red nose is just a sight to see!

[PollenFiend]

Variations of the above pic are now on Pixiv.  Shouldn't need a login to see this one, but yes for other works on my account there.

On 2/19/2023 at 8:59 PM, papabear70 said:

Ohhh, I like this a lot. Her facial expression along with her pinkish red nose is just a sight to see!

Thank you!

[MikiSan]

Have u thought about turning this into an AI chat/rp bot? 

[webmeistro]

5 minutes ago, MikiSan said:

Have u thought about turning this into an AI chat/rp bot? 

AI art has no real connection to AI chatbots, other than the fact that they're what people are talking about now.

You can go to ChatGPT and tell it to pretend it's a girl with hayfever or whatever. It doesn't do horny.

[MikiSan]

6 minutes ago, webmeistro said:

AI art has no real connection to AI chatbots, other than the fact that they're what people are talking about now.

You can go to ChatGPT and tell it to pretend it's a girl with hayfever or whatever. It doesn't do horny.

i dont mean chatgpt I mean AI character. You can literally easily program it to be fetishy with paremeters. 

[mysterysneeze]

I think of the Terminator movies and I laugh thinking of a skynet killer robot suddenly being reprogrammed to having fetishlike personality trait instead of killing...that's a scary but funny scenario.

 

In all seriousness, AI art is evolving. As more people use it the more useful and clear the Ai creates things. I have seen some animators use their skills together with AI to make movement AI art submissions. As far as I know it is possible for an animator's AI program to take a youtube clip or photoshop edit an animator made and the animator could use AI to make something new based on what they already made in the past.

 

https://www.deviantart.com/intiart/art/Busty-Bowsette-Animated-948737000

https://www.deviantart.com/intiart/art/Shy-Marin-Animated-948858066

 

To me personally I know there's a lot of coding and behind the scenes effort and energy that is put into AI programing. It is similar to how google search engines were starting out.

 

I have only spent a few minutes trying Ai art. My results are pitiful because of that. If I spent hours researching and trying different things I know I would get more clear results. It is incredible seeing what other people have figured out with AI art, and I am curious how AI art will develop in the future.

Anyone like me old enough to remember the older searches and loading times inbetween searching when trying to use yahoo and google and msn? XD

 

 

 

 

[MikiSan]

Pfft I managed to make a snz Ai rp bot still adding stuff to it but it learns pretty fast.

[jamesdear1]

That's awesome,! Can you share on how to do that, or share yours

[Spacesneezeman]

AI is actually getting really good at these!

[MikiSan]

On 2/24/2023 at 2:52 PM, Spacesneezeman said:

AI is actually getting really good at these!

I agree made some of mine do a sneezapoolaza.

[PollenFiend]

I'm trying a new model called Deliberate (v2, from https://civitai.com/models/4823/deliberate) and I'm on the fence whether I like this one more or the previous one I've been using which is Realistic Vision 1.3.  What do you all think?  Pic below is my take on my character Lori, also from Start Simulation.  @Chiguy13 prompted me to not forget about this character so I gave it a go.    This is still a work in progress, particularly as I try to make art with the same character and background but in a variety of poses.  Studying up on both ControlNet and DreamBooth tools to go alongside SD.  Might take me a while, so here's a hopeful preview of the character at least.

 

[PapaBear70]

It looks fantastic! She's got a wonderful sneezy look on her face

[Chiguy]

1 hour ago, PollenFiend said:

I'm trying a new model called Deliberate (v2, from https://civitai.com/models/4823/deliberate) and I'm on the fence whether I like this one more or the previous one I've been using which is Realistic Vision 1.3.  What do you all think?  Pic below is my take on my character Lori, also from Start Simulation.  @Chiguy13 prompted me to not forget about this character so I gave it a go.    This is still a work in progress, particularly as I try to make art with the same character and background but in a variety of poses.  Studying up on both ControlNet and DreamBooth tools to go alongside SD.  Might take me a while, so here's a hopeful preview of the character at least.

 

Wow!
I love everything about this, PollenFiend. The face, her pose, the details like the hanky. I really hope you carry on. Just brilliant!

I like both models. I think this makes the Lori character distinct from Chunhua, which is nice for the fans of your stories. 

[baron_von_gotta]

[Spacesneezeman]

On 2/28/2023 at 1:29 AM, PollenFiend said:

I'm trying a new model called Deliberate (v2, from https://civitai.com/models/4823/deliberate) and I'm on the fence whether I like this one more or the previous one I've been using which is Realistic Vision 1.3.  What do you all think?  Pic below is my take on my character Lori, also from Start Simulation.  @Chiguy13 prompted me to not forget about this character so I gave it a go.    This is still a work in progress, particularly as I try to make art with the same character and background but in a variety of poses.  Studying up on both ControlNet and DreamBooth tools to go alongside SD.  Might take me a while, so here's a hopeful preview of the character at least.

 

Love it, AI did another fine job!

[PollenFiend]

On 2/27/2023 at 8:55 PM, Chiguy13 said:

Wow!
I love everything about this, PollenFiend. The face, her pose, the details like the hanky. I really hope you carry on. Just brilliant!

I like both models. I think this makes the Lori character distinct from Chunhua, which is nice for the fans of your stories. 

Hi!  I added two new picture variations and posted them to a standalone thread in the Art board.  I'll leave posts here to talk about tips and lessons learned.  I'm still trying to figure out ControlNet, LORAs, pose editors, etc.

 

On 3/3/2023 at 3:19 PM, baron_von_gotta said:

<pic>

Nice!  The art style is very clean, and I like the detail in the tight shorts.   

 

On 3/18/2023 at 5:32 AM, Spacesneezeman said:

Love it, AI did another fine job!

Thanks!  As mentioned above, I've added two more variation pics and posted them to a standalone thread.

 

Edited March 23, 2023 by PollenFiend