|
     More avatar worthy stuff.
|
#
?
Oct 2, 2022 23:18
|
|
the Mods, they knew!
|
|
| # ? May 31, 2024 19:12 |
|
|
Bula Vinaka posted:Just learned about Loab. Loab was Midjourney, was a minor discord meme a few months before the Twitter thread. I dunno, having spent at least a while using all three of D2, MJ, and SD, I feel like the results of those three are pretty visually distinct. I can usually tell which one a given image came from.
|
|
#
?
Oct 2, 2022 23:46
|
|
|
Can someone
|
|
#
?
Oct 3, 2022 04:07
|
|
|
Hadlock posted:Can someone I’ve spent five minutes on this and can already tell it might be a bitch to get working. I’ll try more tomorrow for funsies.
|
|
#
?
Oct 3, 2022 04:16
|
|
|
Hadlock posted:Can someone It doesn't appear to know what shutter shades are. Here's my best attempt after 5 minutes of work.  Deal with it. It was the 70's. Chewie did a lot of coke, okay? WhiteHowler fucked around with this message at 04:58 on Oct 3, 2022 |
|
#
?
Oct 3, 2022 04:56
|
|
|
I used dreambooth to train a model on the Studio Ghibli style with 300 images. Its really powerful with img2img.      
|
|
#
?
Oct 3, 2022 06:19
|
|
|
IShallRiseAgain posted:I used dreambooth to train a model on the Studio Ghibli style with 300 images. Its really powerful with img2img. was the prompt for this one "windows xp wallpaper"
|
|
#
?
Oct 3, 2022 06:41
|
|
|
Hadlock posted:was the prompt for this one "windows xp wallpaper" I used img2img on it. I actually use the wallpaper for testing img2img prompts. Anyways, here is the model. Prompt is "studio_ghibli_anime_style anime_screencap". (its based on Waifu Diffusion) Here is a bonus image: 
IShallRiseAgain fucked around with this message at 06:58 on Oct 3, 2022 |
|
#
?
Oct 3, 2022 06:54
|
|
|
If you have Stable Diffusion working, is there an easy way to set it up so that you can switch between model files?
|
|
#
?
Oct 3, 2022 06:59
|
|
|
Haven't found a gui option atm. If enough people begin swapping model files it will probably become a thing. Probably as a startup option or something.
|
|
#
?
Oct 3, 2022 07:17
|
|
|
Hadlock posted:Can someone 
|
|
#
?
Oct 3, 2022 07:28
|
|
|
Swedish Thaumocracy posted:Speed for one. "Quality" for another. I can get a Rembrandt in five seconds using Stable Diffusion running on my own GPU, vs spending hours or days toiling in paint and photoshop for a technically inferior result. now that's a good beverage
|
|
#
?
Oct 3, 2022 07:36
|
|
|
 This is both wrong and very wrong skeleton passing his power to a baby under a full moon
|
|
#
?
Oct 3, 2022 08:05
|
|
|
Nessus posted:If you have Stable Diffusion working, is there an easy way to set it up so that you can switch between model files? In the settings menu in automatic1111 (voldy) there’s an option for it You can also merge models
|
|
#
?
Oct 3, 2022 08:06
|
|
|
BARONS CYBER SKULL posted:In the settings menu in automatic1111 (voldy) there’s an option for it
|
|
#
?
Oct 3, 2022 08:23
|
|
|
BARONS CYBER SKULL posted:In the settings menu in automatic1111 (voldy) there’s an option for it Merge, even? Gotta look into that. 
|
|
#
?
Oct 3, 2022 09:27
|
|
|
Mindblast posted:Merge, even? Gotta look into that. 
|
|
#
?
Oct 3, 2022 10:44
|
|
|
What the gently caress. I thought I was relatively up to date. Guess Ill get the latest Automatic1111. Maybe it can autoupdate but I hosed up? Either way; cool poo poo. Gonna look into that tonight.
|
|
#
?
Oct 3, 2022 14:36
|
|
|
Doesn't checkpoint merging carry a significant risk of messing up things in your original "stock" model? Everything I've read about Dreambooth and other training methods says "uh, be VERY sure to back up your existing checkpoint".
|
|
#
?
Oct 3, 2022 14:46
|
|
|
I was having lunch, and a friend mentioned he was watching all of the Alien films again. The rest is history. Ripley having lunch with a Xenomorph from the film Alien, photo realistic, dark and moody  Although it doesn't really look like Ripley (or a Xenomorph), there's obvious cues there. I had the fully grown one in my head - which seems like it would make for a lunch involving more robust dialogue, I'd imagine - but DALL-E had other ideas.
|
|
#
?
Oct 3, 2022 14:57
|
|
|
Mindblast posted:What the gently caress. I thought I was relatively up to date. Guess Ill get the latest Automatic1111. Maybe it can autoupdate but I hosed up? Either way; cool poo poo. Gonna look into that tonight. add the line 'git pull' into the launcher .bat and you get the update every time you launch. sometimes this break it to poo poo.
|
|
#
?
Oct 3, 2022 15:07
|
|
|
SCheeseman posted:Was this trying to make a point, or a joke. Data isn't real btw I sincerely desire a functionless computer. But to say it more clearly, you need time to make something good. If this thing is intelligence, speed should not be a priority. If it is a tool, it should be easier to produce specific results.
|
|
#
?
Oct 3, 2022 15:23
|
|
|
FunkyAl posted:I sincerely desire a functionless computer. But to say it more clearly, you need time to make something good. If this thing is intelligence, speed should not be a priority. If it is a tool, it should be easier to produce specific results. If you told me you were generating every one of your posts in this thread with an AI, I would believe you.
|
|
#
?
Oct 3, 2022 15:43
|
|
|
WhiteHowler posted:If you told me you were generating every one of your posts in this thread with an AI, I would believe you. I am an Al.
|
|
#
?
Oct 3, 2022 17:04
|
|
|
I would really really like them to start making useless computers though. Like a weird shape in a corner that is printing a readout of random numbers on a little receipt paper. Or a computer that will only open your programs if the weather is bad. A computer that intakes all noise it hears and once a day, randomly, makes a noise that is a synthesis of all of them. A computerized toaster that burns toast on purpose. A computer which names every item at the grocery store in order, in case you forgot anything. A computer you need to feed vegetables, and the only thing that it does is generate an ecosystem scene where some animals die if they are not getting enough vegetables. A tamagotchi. That kinda thing.
|
|
#
?
Oct 3, 2022 17:11
|
|
|
|
|
#
?
Oct 3, 2022 17:16
|
|
|
FunkyAl posted:I would really really like them to start making useless computers though. Like a weird shape in a corner that is printing a readout of random numbers on a little receipt paper. Or a computer that will only open your programs if the weather is bad. A computer that intakes all noise it hears and once a day, randomly, makes a noise that is a synthesis of all of them. A computerized toaster that burns toast on purpose. A computer which names every item at the grocery store in order, in case you forgot anything. A computer you need to feed vegetables, and the only thing that it does is generate an ecosystem scene where some animals die if they are not getting enough vegetables. A tamagotchi. That kinda thing. loading joker.exe...
|
|
#
?
Oct 3, 2022 17:18
|
|
|
FunkyAl posted:I would really really like them to start making useless computers though. Like a weird shape in a corner that is printing a readout of random numbers on a little receipt paper. Or a computer that will only open your programs if the weather is bad. A computer that intakes all noise it hears and once a day, randomly, makes a noise that is a synthesis of all of them. A computerized toaster that burns toast on purpose. A computer which names every item at the grocery store in order, in case you forgot anything. A computer you need to feed vegetables, and the only thing that it does is generate an ecosystem scene where some animals die if they are not getting enough vegetables. A tamagotchi. That kinda thing. I think what you’re describing is called “art”, and the medium is something called a “computer”. hth
|
|
#
?
Oct 3, 2022 17:24
|
|
|
dumb question -- what's the model's name where you can prompt it with a landscape with some text, where the generation will be imagery placing the object listed in the corresponding location on the page. So you can determine which quadrant of a landscape has a river vs castle, etc. I know I've seen this done, but I can't remember the right term for it.
|
|
#
?
Oct 3, 2022 17:51
|
|
|
DALL-E I think can insert an object wherever you want to. Or maybe it's just the image2image function in SD and I assume others? You can MS-Paint a landscape and have it convert it into a painting or something 
|
|
#
?
Oct 3, 2022 18:11
|
|
|
Not quite the same thing -- imagine if the MS-Paint example you are showing didn't have any drawn imagery but was more a powerpoint-caption thing. It would just have text centered in the various locations: "sand," "grass," "river," "tree", "mountain" without any lines or color. The model would take that text AS the text prompt, without any additional overall text prompting, and generate the coherent whole that had components that roughly matched the locations (and meanings) for those words. It might be what you're referring to for Dalle-2, but I'm having trouble pulling up examples for it. I've seen it mostly for landscapes, but I don't think it's only used for that.
|
|
#
?
Oct 3, 2022 18:35
|
|
|
madmatt112 posted:I think what you’re describing is called “art”, and the medium is something called a “computer”. hth Yes. People use computers to make art all the time. The ai art does not make sense because it is a program bloated with 10,000 gigs of images, trying to replicate approaches to painting that were popular 100 years ago. It's using tools to make things needlessly complicated.
|
|
#
?
Oct 3, 2022 18:41
|
|
|
|
|
#
?
Oct 3, 2022 18:43
|
|
|
"Art isn't art when the procedure used to create it is needlessly complicated" is the kind of stupid-rear end observation I would expect from unfiltered AI generated text output but I guess that's at least in keeping with your gimmick.
|
|
#
?
Oct 3, 2022 18:49
|
|
|
mobby_6kl posted:DALL-E I think can insert an object wherever you want to. I figured out the term I was missing: "layout-to-image." Really seems to be way less frequently used with less recent progress than other techniques. Curious whether it'll see more attention or not, as it looks like it would give more control over how to compose an image than the single descriptive prompt method. Though the image2image with a sketch gets there too. Example: 
|
|
#
?
Oct 3, 2022 18:52
|
|
|
jemand posted:I figured out the term I was missing: "layout-to-image." Really seems to be way less frequently used with less recent progress than other techniques. Curious whether it'll see more attention or not, as it looks like it would give more control over how to compose an image than the single descriptive prompt method. Though the image2image with a sketch gets there too. Huh, thanks for finding it, I don't remember seeing this before. But it totally makes sense since the algorithm would have to find where to best place objects anyway.
|
|
#
?
Oct 3, 2022 19:02
|
|
|
Skeletons throwing a surprise birthday party for Dracula in the style of a polaroid photo   
|
|
#
?
Oct 3, 2022 19:11
|
|
|
Stable Diffusion can't draw a starship Enterprise (no bloody A, B, C, or D -- but those either). I've tried several keywords (starship, Enterprise, NCC-1701, Star Trek, etc.). This is the closest it's gotten in about 30 tries with varying CFG values:  There should be plenty of source images on the Internet, so that's rather weird -- especially since SD can passably draw any given Star Trek character you ask for.    But no Enterprise. Truly a massive gap in the AI's modeling.
|
|
#
?
Oct 3, 2022 19:58
|
|
|
SCheeseman posted:"Art isn't art when the procedure used to create it is needlessly complicated" is the kind of stupid-rear end observation I would expect from unfiltered AI generated text output but I guess that's at least in keeping with your gimmick. Usually a procedure this complex produces better art.
|
|
#
?
Oct 3, 2022 20:05
|
|
|
|
| # ? May 31, 2024 19:12 |
|
|
|
|
#
?
Oct 3, 2022 20:15
|
|