I'm also currently playing around with StableDiffusion. Got it running today and only play around with text2image atm. Want to get into image2image on the WE and then do photorealistic Marseys.
Not Marsey, but here's "anime of a capybara family taking a bath":
I think our prompts just suck. Or the amount of generated images ;)
I checked out /r/StableDiffusion/ and they do really have some elaborate prompts, so there's probably a lot to dive into.
This post also described their workflow. Generate tons of images and then use the ones they like the most as initial seed for further iterations (that is where the img2img comes into play).
Wait, how'd you get that running on your own PC? Do you not have to download the 50 TB dataset to use it? I tried looking at it yesterday and I was lead to believe that you need an obscenely large dataset for local usage
Many reddit and 4chan threads later, I've learned that people with GTX 16xx cards seem to all be getting the green images. Using a fix for this (or doing what I did, following the updated r-slur guide) just causes full vram and then a crash
no one has posted any solutions for us 16xxcels and I need to sleep and not figure this out
Relatively easy. I followed an old version of this guide.
Keep in mind though that it also depends on your hardware (Nvidia graphics card is easy - except another dramatard mentioned an issue with a specific model, AMD is harder).
Jump in the discussion.
No email address required.
I'm also currently playing around with StableDiffusion. Got it running today and only play around with text2image atm. Want to get into image2image on the WE and then do photorealistic Marseys.
Not Marsey, but here's "anime of a capybara family taking a bath":
Jump in the discussion.
No email address required.
I've been messing with Stable Diffusion as well with some mixed results. These were probably some of my best results
Jump in the discussion.
No email address required.
I think our prompts just suck. Or the amount of generated images ;)
I checked out /r/StableDiffusion/ and they do really have some elaborate prompts, so there's probably a lot to dive into.
This post also described their workflow. Generate tons of images and then use the ones they like the most as initial seed for further iterations (that is where the img2img comes into play).
Jump in the discussion.
No email address required.
jesus @ that prompt. I'll need an GPT bot to analyze the prompts and generate them for me.
Jump in the discussion.
No email address required.
Prompt: dramamine bussy, dramamine boy kitty, dramamine butthole, dramamine starfish, dramamine anus, dramamine rectum, dramamine butthole, dramamine sphinter, !gussy
Jump in the discussion.
No email address required.
You don't want to see the results...
or do you
Jump in the discussion.
No email address required.
@Dramamine accurate?
Jump in the discussion.
No email address required.
looks cursed by God himself, so probably
Jump in the discussion.
No email address required.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context
One day I'm going to find you and we're going to fist fight and have some beers afterwards. I don't like that this is the reality, but that's just it.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context
More options
Context
I raise your capybara family with capybara gaming
Jump in the discussion.
No email address required.
@Aevann
Jump in the discussion.
No email address required.
More options
Context
More options
Context
Wait, how'd you get that running on your own PC? Do you not have to download the 50 TB dataset to use it? I tried looking at it yesterday and I was lead to believe that you need an obscenely large dataset for local usage
Jump in the discussion.
No email address required.
Nah, you only need to download the finished model (~4gb) and some additional python stuff (didn't want to mess with my own python installation).
Basically I followed this aptly named guide.
Jump in the discussion.
No email address required.
I did the guide exactly and got all green images
mfw too r-slurred for the r-slur guide
edit:
that guide is out of date apparently?
updated r-slur guide
I will see if this one works
edit2:
Many reddit and 4chan threads later, I've learned that people with GTX 16xx cards seem to all be getting the green images. Using a fix for this (or doing what I did, following the updated r-slur guide) just causes full vram and then a crash
no one has posted any solutions for us 16xxcels and I need to sleep and not figure this out
Jump in the discussion.
No email address required.
It's telling you to restore Gaddafi to power.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
Ngmi
Jump in the discussion.
No email address required.
More options
Context
More options
Context
Relatively easy. I followed an old version of this guide.
Keep in mind though that it also depends on your hardware (Nvidia graphics card is easy - except another dramatard mentioned an issue with a specific model, AMD is harder).
Jump in the discussion.
No email address required.
More options
Context
More options
Context