AI [AI Art] - Show Us Your AI Skill *NO TEENS*

[Mr-Fox]

Great job. However, I believe that you need to do some adjustments to your settings. When you get artifacts and/or everything looks a bit overly sharp you need to mainly look over the sampling steps but also the denoising strength if you are using Highres fix or upscaling. I would also highly recommend you trying out different checkpoints and samplers with the same prompt and seed. Plot script is great for this.

I made the last post before actually taking a look at the meta data. Now that I have. You have done a lot of good work with your prompt and settings. It only requires a few smal tweaks to take it to the next level. Always use "text, logo, watermark," in negative prompt, you might also need to weight them. A major issue, never use SwinIR as upscaler. Especially not in upscaling in either img2img or extra tab. It will remove all texture and a lot of detail fidelity. Instead use Lanczos for upscaling.
You had used a high amount of sample steps and low Hires steps, invert this. Use 20-30 sample steps and 40-60 Hires steps, I know it sucks how long it takes.
After this you can play around a bit with denoising strength but I would recommend ap 0.3 .
In my experience you get better result by not using Face restoration with Hires fix, instead use post processing.
I recommend using GFPGAN.
If you are struggling with artifacts and/or distortion such as fingernails etc, clip skip 2 can do wonders sometimes.
Read more about it

You must be registered to see the links

.
Don't forget to use negative prompting "bad fingernails" "ugly fingernails" in combination with "detailed fingernails" in positive etc.

 

[Mr-Fox]

This is the awesome prompt by our very own fr1ko, I only tweaked it a little.
I started out recreating the same image with the same settings and achieved an exact replica, Then I started to tweak the settings in accordance to my own recommendations.
After a few generations and more tweaking:

I wanted to demonstrate improved hands and nails but SD hid them instead.
Like a big middle finger for this "smartypants".. Well played..
Next I wanted to try bring back the original concept and theme more so I gave those tags a bit of weight.
E Voilá, Goddess:


There are of course room left for more improvements but I leave that to the eminent op fr1ko.

 

[rayminator]

 

[rayminator]

some more Sonic and Amy Rose

guessing that sonic is really a female every time I put him in he always a female

I'm using Sonic Diffusion v3 beta

Prompts:

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

Negative:

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

 

[Telgar]

Some of my more detailed works.

 

[modine2021]

 

[modine2021]

 

[modine2021]

 

[sharlotte]

Going for brunettes this time.

 

[atheran]

Not gonna bother editing those.



vs



Which is better? A blind idiot like me, can't see much change.
First one is with UltraSharp upscaler, second is with foolhardy_remacri. Everything else is exactly the same, including highres steps etc.

And a cutie after the experiment. Also not processed.

 

[BlackBarba]

Playing with Artuis v2

 

[Mr-Fox]

Not gonna bother editing those.

View attachment 2509510

vs

View attachment 2509512

Which is better? A blind idiot like me, can't see much change.
First one is with UltraSharp upscaler, second is with foolhardy_remacri. Everything else is exactly the same, including highres steps etc.

And a cutie after the experiment. Also not processed.

View attachment 2509515

Ultrasharp has a tiny bit more vibrance/saturation or possibly just a bit more orange bias in the color palette. you can see it in the hair. Possibly a tiny bit more exposure if you look at the bright spots in the hair. foolhardy_remacri has a bit more clarity with the details and shadows being slightly darker. I think that ultrasharp is great for semirealism and colorful images. I have not tried foolhardy but it looks aimed towards photo realism and possibly noir and artistic photos.
Just my 2 freedom pennies..

 

[fr34ky]

Totally reliable bride Jinx

 

[daddyCzapo]

This is the awesome prompt by our very own fr1ko, I only tweaked it a little.
I started out recreating the same image with the same settings and achieved an exact replica, Then I started to tweak the settings in accordance to my own recommendations.
After a few generations and more tweaking:
View attachment 2507540
I wanted to demonstrate improved hands and nails but SD hid them instead.
Like a big middle finger for this "smartypants".. Well played..
Next I wanted to try bring back the original concept and theme more so I gave those tags a bit of weight.
E Voilá, Goddess:
View attachment 2507542

There are of course room left for more improvements but I leave that to the eminent op fr1ko.

Thank you for the advice on everything! <3 However i fucked myself over thinking i can pull that off with my setup. Turns out that Lanczos clogs the vram so much that my pc had a few bluescreens and then i had to reinstall webui completely... something went sideways as always XD That SwinIR upscaler was a whim of the moment, normally i use 8x NMKD Superscale in hires as well as img2img/extras. I'll definetely try to perfect that line of prompts, but not at the moment

 

[Mr-Fox]

Thank you for the advice on everything! <3 However i fucked myself over thinking i can pull that off with my setup. Turns out that Lanczos clogs the vram so much that my pc had a few bluescreens and then i had to reinstall webui completely... something went sideways as always XD That SwinIR upscaler was a whim of the moment, normally i use 8x NMKD Superscale in hires as well as img2img/extras. I'll definetely try to perfect that line of prompts, but not at the moment

Oh man that sucks. It's strange that you had vram problem with Lanczos, it's always much faster than SwinIR on my machine.

 

[RedRidingWolf]

Is there a good AI generator for free to test?

 

[BlackBarba]

Is there a good AI generator for free to test?

There are the famous Discord ones like

You must be registered to see the links

and

You must be registered to see the links

, but I don't think you can create NSFW stuff there.

Another option would be to use

You must be registered to see the links

which is a crowdsourced cluster, in short people provide their GPU power and you can use it "for free". It uses Kudos as currency to increase its priority in render queues. But that doesn't stop you from using it for free, you don't even need to create an account, it can just be a little slow.
The advantage is that you have access to all the fancy stuff, lots of models, inpainting, img2img, and even ControlNet (on its official client), and most importantly, you can create NSFW stuff.

Stable horde is just an "API", you need a client to run it, here is few of them:

You must be registered to see the links

(their official client)

You must be registered to see the links

You must be registered to see the links

Last but most powerful option is to use Google Colab.
Did you know that Google provides GPU power for you to use for free? It has some limitations like, you can only 12 hours a day (just switch to another account when that happens). It's a little more complex to get started, but it's worth it. Because you will have access to all the possibilities of Stable Diffusion. Like, found a new model/lora model? Just upload it to your Google Drive and start using it. There is a update in ControlNet? Just download it and use it.

There are 2 main disadvantages:
1 - It will take up a lot of space in your Google Drive, especially if you start downloading a lot of models, loras, etc. 15 GB is not going to be enough.
2 - Google Collab should be used for serious stuff like research, I don't know anyone who got banned for creating waifus all day, but keep that in mind.

There are several ways to use Stable Diffusion in Google Colab, but I think this is the easiest one:

You must be registered to see the links

 

[RedRidingWolf]

Thanks, now I can try to create my next D&D Character

 

[PandaRepublic]

 

[rayminator]

here some more all of my image made ai or other I always add the ages above 18

santa:


a mistake


better



Futa