Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

[Psicopax]

been working on this today after ruining Stable Diffusion trying to update an extension. Took like 5 hours to figure out what i fucked up. But fixed it in the end.


Some different version. I like the black and white ones the most. Has that manga look.

 

[Abeezo]

more stuff, haven't been keeping up with AI news as of late so reading the thread is a good way to catch up!

 

[anustart76]

Decided to try creating some images with "Terran Trade Authority" in the prompt. The results were a pleasant surprise. Hope some of the rest of you enjoy these.

View attachment 4978125View attachment 4978129

That blue and purple one is giving me serious nostalgia for the level of Marvel Ultimate Alliance where you fight Galactus on the Skrull homeworld.

 

[purpleflurp]

been working on this today after ruining Stable Diffusion trying to update an extension. Took like 5 hours to figure out what i fucked up. But fixed it in the end.

View attachment 4989897View attachment 4989893View attachment 4989905
Some different version. I like the black and white ones the most. Has that manga look.

That looks really similar to the model I've been building. What are you using for the dataset?

 

[Psicopax]

That looks really similar to the model I've been building. What are you using for the dataset?
View attachment 5001709View attachment 5001713

I'm mostly using Loras to edit existing models as i'm not too well versed in creating my own model yet.
But using Loras make it so i can edit small values and get exactly what i want. Like getting this sort of manga style with a mix of ancient and modern creature and building design.

Model: AEPASTMIX (I can't find this model online anywhere and i dont remember where i got it. If there is request for it i can probably upload it)

Lora: epinoiseoffset, danmumfordStyleLora, animeLineartMangaLike.
Embedding/Textual Inversion - Positive: rfktr_bwmnga
Embedding/Textual Inversion - Negative: bad-hands-5, easynegative

Spoiler: NSFW

 

[Abeezo]

I've been playing around with this model called MIX-Pro-V4, might be my main model I like it alot so far highly recommend

 

[WelfareNiggerQueen]

I've been playing around with this model called MIX-Pro-V4, might be my main model I like it alot so far highly recommend

For me, It's Barron's mix. It's the only one that works with my autistic prompting style.

 

[Balr0g]

Happy Easter fellow Kiwis

Edit: Reuploaded

 

[Agent of Z.O.G.]

View attachment 4916445
I was playing with image 2 image using this picture of cobra. Some nice results but they all made him look better than he already is as opposed to worse. The funniest part is some of these generations makes him look like a back alley street merchant.

Results:
View attachment 4916382View attachment 4916386View attachment 4916390View attachment 4916394View attachment 4916398View attachment 4916410

Midjourney cut me off since they're not allowing trial accounts at the moment, do I have to finally learn how to use stable diffusion for my img2img needs or is there something easier for a lazy fuck like me?

 

[Psicopax]

Happy Easter fellow Kiwis

View attachment 5015713

Happy easter friend! Try uploading the pic again? Doesn’t seem like it worked the first time.

Midjourney cut me off since they're not allowing trial accounts at the moment, do I have to finally learn how to use stable diffusion for my img2img needs or is there something easier for a lazy fuck like me?

You can try some other online ones. I know Google Colab has a system. Tho Stable Diffusion isn’t that difficult to install anymore. You could try your luck with this easy installer.

If you choose to install it, the Automatic 11111 or whatever is the one you want. It has a slick web ui

 

[WelfareNiggerQueen]

Happy Easter fellow Kiwis

View attachment 5021362

Edit: Reuploaded

You need some vae, nigger. https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main

 

[Balr0g]

You need some vae, nigger. https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main

Have to get back into it I stayed away from open diffusion for a while. Thx for the info

 

[Gamepads]

Have to get back into it I stayed away from open diffusion for a while. Thx for the info

To clarify for anyone else reading this thread: No VAEs or improper VAE with your model usually results in washed out outputs, and occasional black or purple splotches or smudges.

There are models that have their own VAE, but that still requires you to download them as they'll be separate files.

 

[Psicopax]

To clarify for anyone else reading this thread: No VAEs or improper VAE with your model usually results in washed out outputs, and occasional black or purple splotches or smudges.

There are models that have their own VAE, but that still requires you to download them as they'll be separate files.

What I’m working on today.

Tbh I just downloaded all VAE I came across and put them in my VAE folder with SymbolicLinks to them in my Models folder. Then I just pick Automatic as my VAE in the settings. Usually picks the right one automatically. But then again I have 700gb of models and Loras and can’t keep track of it all. I only change from Auto if I notice anything wrong.

SymbolicLink is worth reading about btw,

Spoiler: What Is Symbolic Link?

if anyone don’t know what it is. Since I learned about it I find great uses for it all the time. In simple terms, it’s a more advanced shortcut where software believes it is the real file, but it’s just being forwarded to the real location of the file. You can do it with any folder and any file. So I have all my models on an external drive and have just made symbolic links to them in the Stable Diffusion models folder. Works like a charm and doesn’t waste my C drive.

 

[Abeezo]

Some more stuff but I'm experimenting with some img2img and wow the results were pretty good!

 

[purpleflurp]

But then again I have 700gb of models and Loras and can’t keep track of it all. I only change from Auto if I notice anything wrong.

I've been in a constant state of buttmad because of Paperspace's 50gb storage cap. For now I can only dream of the fuckery I'll be doing once I'm not limited to the storage capacity of a fucking flash stick

 

[Psicopax]

I've been in a constant state of buttmad because of Paperspace's 50gb storage cap. For now I can only dream of the fuckery I'll be doing once I'm not limited to the storage capacity of a fucking flash stick

I feel your pain. I just recently bought a new HDD because I got so sick of trying to minmax my hard drive space. I can’t imagine how annoying it would be to have to deal with that, but with a 50gb limit.

 

[purpleflurp]

I feel your pain. I just recently bought a new HDD because I got so sick of trying to minmax my hard drive space. I can’t imagine how annoying it would be to have to deal with that, but with a 50gb limit.

Ordinarily I'm fine with small storage sizes, after all I grew up in the time when 512mb flash sticks were starting to replace floppies. But when you're dealing with individual files that are 4gb+, it turns into an annoying game of tetris as I swap checkpoints between FTP servers.

The main shit thing is they don't actually cap you off at 50gb, nor is there an option to do so. You can very easily go over the 50gb limit if you have multiple notebooks, after which they charge you per gigabyte. But they don't tell you how much you owe until the 1st of the month, which can lead to a surprise $60-70 bill.

It got to the point where I was considering upgrading to their enterprise plan, since that tier bypasses storage limits and the 6 hour session cap. But that would require me shelling another ~$400 to incorporate an LLC on top of whatever monthly/annual rates the enterprise plan would cost me. By that point the money's better invested into slapping an Enterprise GPU into my workstation.

Apologies for the autistic sneeding about cloud computing, they've just been riling me up for a while.

 

[Girl Named Sandoz]

Ordinarily I'm fine with small storage sizes, after all I grew up in the time when 512mb flash sticks were slowly starting to replace floppies. But when you're dealing with individual files that are 4gb+, it turns into an annoying game of tetris mixed in with swapping files between FTP servers.

if you aren't already, you should try pruning your models and converting to fp16. this can reduce them down to around 2GB each and the quality is indistinguishable from the full precision models. if you're on webUI there's even a handy extension that will do it for you, in addition to things like fixing broken clip

 

[Psicopax]

Ordinarily I'm fine with small storage sizes, after all I grew up in the time when 512mb flash sticks were starting to replace floppies. But when you're dealing with individual files that are 4gb+, it turns into an annoying game of tetris as I swap checkpoints between FTP servers.

The main shit thing is they don't actually cap you off at 50gb, nor is there an option to do so. You can very easily go over the 50gb limit if you have multiple notebooks, after which they charge you per gigabyte. But they don't tell you how much you owe until the 1st of the month, which can lead to a surprise $60-70 bill.

It got to the point where I was considering upgrading to their enterprise plan, since that tier bypasses storage limits and the 6 hour session cap. But that would require me shelling another ~$400 to incorporate an LLC on top of whatever monthly/annual rates the enterprise plan would cost me. By that point the money's better invested into slapping an Enterprise GPU into my workstation.

Apologies for the autistic sneeding about cloud computing, they've just been riling me up for a while.

Is there a specific reason you are not running it locally? From my testing you can basically run it on GPU cards that are 2 generations old. With obvious speed setbacks. But manageable.

Personally I’m running Stable Diffusion fine on my pc with a 2080ti. I run into some issues with GPU memory at times but I’ve been working on finding ways to optimize it. so if you have a workstation or computer with around the specs of a 2080, it works just fine. Probably isn’t as fast as a dedicated cloud computer tho.

As Sandoz says here:

if you aren't already, you should try pruning your models and converting to fp16. this can reduce them down to around 2GB each and the quality is indistinguishable from the full precision models. if you're on webUI there's even a handy extension that will do it for you, in addition to things like fixing broken clip

If you are not planning on combining models to make your own models or to train your own model using these models, then getting fp16 models basically has no drawback and it can cut the size in half.

Many people will upload fp16 versions of their models now, and often respond to requests to do so in the comment section, especially on Civitai. But be sure you don’t wanna use them for training. As fb16 versions do not have the training data in them.

I’ve started downloading all the models and loras I want to try because a lot are getting removed. TOS changes, legal issues, artists getting mad and demanding it taken down, FUCKING EXCLUSIVITY DEALS where they sell the model to AI generating websites and remove it everywhere like what happened recently with the Illuminati model (the cocksuckers ) etc. So I download everything I think looks cool, save a PDF of the download page so I have info about triggers and best settings, and then store it away.