|
My favorite result so far, SD using the web demo: Prompt: full color photograph of a border collie at a DJ turntable with fireworks and laser lights, vibrant, colorful, sharp
|
#
?
Sep 7, 2022 23:38
|
|
|
|
| # ? May 26, 2024 17:24 |
|
|
If this came out last year I could have made SO MUCH money off complete idiots with stupid loving NFTs click for absolutely loving huge
|
|
#
?
Sep 8, 2022 00:03
|
|
|
WhiteHowler posted:The image in the upper right is cool as heck. you should buy a midjourny subscription. its $30 per month for unlimited renders in slow mode. I am making DM material as we speak, in the background! i've only had it a few days and i've made at least 1200 things 
|
|
#
?
Sep 8, 2022 00:45
|
|
|
Just playing around with MJ:  
|
|
#
?
Sep 8, 2022 00:59
|
|
|
i've been having good results with midjourney asking for vague things and single word prompts time  god  power  youth  age  a life well lived 
|
|
#
?
Sep 8, 2022 01:24
|
|
|
Rutibex posted:i've been having good results with midjourney asking for vague things and single word prompts seems it has a color palette it likes to use
|
|
#
?
Sep 8, 2022 01:49
|
|
|
Rutibex posted:you should buy a midjourny subscription. its $30 per month for unlimited renders in slow mode. I am making DM material as we speak, in the background! i've only had it a few days and i've made at least 1200 things Oh, nice. That sounds worth it for a month
|
|
#
?
Sep 8, 2022 02:00
|
|
|
New Waifu diffusion just dropped for SD https://storage.googleapis.com/ws-store2/wd-v1-2-full-ema.ckpt
|
|
#
?
Sep 8, 2022 03:58
|
|
|
Why in all hell is MJ using discord for front end. I don't want social scrolling noise of everybody elses' stupid projects. I just want a prompt for my things, goddamit.
|
|
#
?
Sep 8, 2022 05:14
|
|
|
 crowd surfing and stage diving at the pennywise show at warped tour 2003, stage photo, bluray still
|
|
#
?
Sep 8, 2022 05:18
|
|
|
frumpykvetchbot posted:Why in all hell is MJ using discord for front end. I don't want social scrolling noise of everybody elses' stupid projects. I just want a prompt for my things, goddamit. You can invite the bot into a private chat.
|
|
#
?
Sep 8, 2022 05:19
|
|
|
frumpykvetchbot posted:Why in all hell is MJ using discord for front end. I don't want social scrolling noise of everybody elses' stupid projects. I just want a prompt for my things, goddamit. to incentivize people to pay for premium
|
|
#
?
Sep 8, 2022 05:26
|
|
|
Boba Pearl posted:New Waifu diffusion just dropped for SD doesn't seem that different from the normal model
|
|
#
?
Sep 8, 2022 05:51
|
|
|
Tunicate posted:to incentivize people to pay for premium Sure, but even their top tier $50/month (billed annually) "enterprise" plan still goes through discord, which is gibberish to me. Is this what getting old is like? I'm like not into compulsory social experiences.
|
|
#
?
Sep 8, 2022 05:52
|
|
|
Tunicate posted:doesn't seem that different from the normal model It seems lot more coherent for anime stuff. Same seed, "son goku driving a car" Waifu:  SD:  But it makes everything anime: Same seed, "Formal portrait of Barack Obama" Waifu:  SD: 
|
|
#
?
Sep 8, 2022 06:11
|
|
|
lunar detritus posted:It seems lot more coherent for anime stuff. Yeah I got a whole lot of this for anything Dragon Ball, I just kinda assumed it might have been trained on all the terrible DB fanart out there. I tried to do a fusion between Goku and Alf though and got some cool stuff    Then I did one starting from a picture of myself with just "Saiyan" for the phrase. Had to adjust the strength a lot for it, and 99% of the results were garbage but   Now I've been trying to do human chimpanzee hybrids for the last half hour but it just keeps giving me normal chimps with boobs/weird nipples :/
|
|
#
?
Sep 8, 2022 06:25
|
|
|
seems like going to safebooru and stealing a list of tags works decently with the anime model  Tunicate fucked around with this message at 07:10 on Sep 8, 2022 |
|
#
?
Sep 8, 2022 07:03
|
|
|
BrainDance posted:Could anyone explain the general principle behind how stable diffusion works? But not in an incredibly technical way because I'm only so smart. If you take an image and then add a little bit of noise to it, you get a slightly noisier, but recognizable picture. Now you do it again 500 times and you probably have an image that is just noise. (arrow from left to right)  What if you start with noise and modify it just a little bit and try to make it look like a cat? (arrow from right to left) There is one neural net called CLIP that looks at images and gives them a textual description. It basically goes "My best guess is that this is 'a picture of a cat lying on a table.' I'm 9% sure" The other neural net then tries to modify the noisy image and try to make the percentage as high as possible. The two neural nets have been trained with millions of image-description pairs on thousands of machines for hundreds of hours to get them to be good at it. It is very expensive. That is the "not incredibly technical way" of describing the process.
|
|
#
?
Sep 8, 2022 07:52
|
|
|
Interestingly, given the amount of classical art that was clearly in the training set, and ironically given the name, Stable Diffusion doesn't seem to really know what a centaur is. It seems to give you horses, or maybe people on horses. (The weeb version is even worse, it just gives you people; but it wants everything to be an anime girl so I'm not surprised) Brutal Garcon fucked around with this message at 09:37 on Sep 8, 2022 |
|
#
?
Sep 8, 2022 09:35
|
|
|
Dall-e 2 knows what a "succubus" is and apparently it's too sexy for me. I must be satisfied with simple "demon girls" 
|
|
#
?
Sep 8, 2022 09:37
|
|
|
Playing around with the weeb model. It kinda knows who "Kujo Jotaro" is, in that it gave me a handsome angular dark-haired man. I also said he should have a hat, which it has taken to heart. 
|
|
#
?
Sep 8, 2022 09:53
|
|
|
So, I was trying to go for a field of crucified mecha in SD, but I got crucifix mecha instead. "Field of crucified japanese mecha, Hyper Detail, 8K, HD, Octane Rendering, Unreal Engine, V-Ray, full hd"  Switched it up to Hello Kitty characters and got something weird and delightful and kinda messed up when you look in the background. "Field of crucified hello kitty characters, Hyper Detail, 8K, HD, Octane Rendering, Unreal Engine, V-Ray, full hd"  Golgotha has never looked so cute. Vlaphor fucked around with this message at 14:02 on Sep 8, 2022 |
|
#
?
Sep 8, 2022 13:53
|
|
|
Wheany posted:If you take an image and then add a little bit of noise to it, you get a slightly noisier, but recognizable picture. Now you do it again 500 times and you probably have an image that is just noise. (arrow from left to right) Basically we've trained a machine to imagine by teaching it to forget in reverse
|
|
#
?
Sep 8, 2022 17:01
|
|
|
First version of a SD photoshop plugin just launched. https://t.co/gqFWpABQLY Haven't tested it myself yet. Getting the feeling this doesn't use your own gpu. But still!
|
|
#
?
Sep 8, 2022 17:34
|
|
|
Mindblast posted:First version of a SD photoshop plugin just launched. The discord says it uses Dream Studio API, which I'm not sure supports inpainting this might just be you can quickly load it make modifications and send back to img2img. I do usually have GIMP open when working in SD locally so it's not a horrible idea but I'd much prefer a local and full featured plugin (and for GIMP I'm not paying Adobe every month for a hobby)
|
|
#
?
Sep 8, 2022 17:37
|
|
|
Mindblast posted:First version of a SD photoshop plugin just launched. Yeah it's using the dream studio API, so the rendering is cloud-based. Probably the easiest way to get going.
|
|
#
?
Sep 8, 2022 17:40
|
|
|
Having fun rendering raccoons and trashcans in the style of famous artists Van Gogh    This one is Thomas Kinkade 
|
|
#
?
Sep 8, 2022 17:57
|
|
|
And here is a Thomas Kinkade McMansion
|
|
#
?
Sep 8, 2022 17:59
|
|
|
pixaal posted:The discord says it uses Dream Studio API, which I'm not sure supports inpainting this might just be you can quickly load it make modifications and send back to img2img. I do usually have GIMP open when working in SD locally so it's not a horrible idea but I'd much prefer a local and full featured plugin (and for GIMP I'm not paying Adobe every month for a hobby) What about Krita? https://github.com/sddebz/stable-diffusion-krita-plugin
|
|
#
?
Sep 8, 2022 18:38
|
|
|
Boba Pearl posted:What about Krita? This is the real, true way. If you're died in the wool photoshop it isn't that difficult to get going with Krita only to have Stable Diffusion running locally on your own machine in a true image editing suite. That Photoshop plugin looks great but burning your Dreamstudio credits blindly in Photoshop seems bad.
|
|
#
?
Sep 8, 2022 18:59
|
|
|
In honour of the news, this is the best I could do with SD without too much effort.
|
|
#
?
Sep 8, 2022 19:26
|
|
|
Freddy Kruger in the style of Lisa Frank -- Midjourney 
|
|
#
?
Sep 8, 2022 19:41
|
|
|
New Trinart tuning just dropped more anime: https://huggingface.co/naclbit/trinart_stable_diffusion_v2
Boba Pearl fucked around with this message at 23:05 on Sep 8, 2022 |
|
#
?
Sep 8, 2022 20:00
|
|
|
Burning Rain posted:In honour of the news, this is the best I could do with SD without too much effort.     Stable Diffusion posted:her undead royal highness queen elizabeth smiling,
|
|
#
?
Sep 8, 2022 20:15
|
|
|
 i just did 'zombie Queen Elizabeth rises from the grave, 4k blu ray still --testp' in midjourney
|
|
#
?
Sep 8, 2022 20:22
|
|
|
We doing a creepy queen theme tonight? I'm in when I'm home
|
|
#
?
Sep 8, 2022 20:23
|
|
|
Humbug Scoolbus posted:Freddy Kruger in the style of Lisa Frank -- Midjourney
|
|
#
?
Sep 8, 2022 20:37
|
|
|
Guess I'll just quote my own post from ~a week ago (in the UK megathread):WhatEvil posted:Queen Elizabeth II the eternal lich, sitting on her throne of skulls:
|
|
#
?
Sep 8, 2022 20:43
|
|
|
detailed photo of an old bronze patina statue of undead Queen Elizabeth II, photorealism, intricate detail, museum diffuse lighting
|
|
#
?
Sep 8, 2022 21:23
|
|
|
|
| # ? May 26, 2024 17:24 |
|
|
  
|
|
#
?
Sep 8, 2022 21:58
|
|