|
If we're going to analogize this stuff at all it's more like throwing clay in that you have a lump of 'stuff' that you chuck at a wheel and start spinning. Most artists I have known usually don't have a specific idea in their head when they start, just a general idea (a cup, vase, something for mothers' day, a pencil holder) and you work it till something forms. Sometimes it seems like everything you throw is amazing, sometimes it seems like nothing you do is making anything and it all sucks. Prompting is really similar to me, you throw words and sometimes amazing stuff comes out right away, sometimes it's all bullshit and nothing is working no matter how much you work on it. MOST of the time you throw something, an output sparks something, and you work toward whatever it suggested to you. For every picture or sound clip I post there are tens of others that didn't work with a whole heap of words that lead nowhere. It's the same with writing, sometimes you're on fire and everything you write is gold, other times nothing is forming and you are having a poo poo day and can't write a goddamn. Anyone that's ever created without AI knows that you can spend an hour on whatever, an image, song, tweaking and then just go "no, this sucks" and just drop it and flush all that effort away because it ain't working. Using these AI systems, and I know I've said this about myself more than a few times this year (almost literally, June 1st 2022 is when I got midjourey access early on) fits exactly in the same headspace as all the other creative poo poo I've ever done. Just now you get to collaborate with a dumb artless robot but boy it's eager to do whatever you want whenever you want without complaining. Bucnasti posted:That baby is rad as hell and I would totally party with him. "A circle logo with a chubby baby glam rock god and bright colors on it, in the style of dark compositions, cypherpunk, linear illustrations, layered colors, poolcore, emphasis on character design" EDIT And to add to the clay throwing analogy, if you put that prompt in Midjourney most of the time you'll get straight-on shots of the baby looking right at you, but if you add "side profile" in the prompt suddenly you get something more like that baby I originally posted, without 'side profile' and with:   I have no idea what 'poolcore' is but when you ask MidJourney what a picture is with the /describe function it ends up suggesting a lot of ----core words so it must mean something to it. This is also where the clay analogy comes into it even more, I don't know all the compositions of clays and dirts and all that when sitting at a clay wheel, but I do know how to manipulate them to get neat results. This is why words that don't seem to mean much make their way into whatever prompts I fire at these systems because it doesn't matter that much. Think of it like seasoning, 'poolcore' means something to the AI so it changes the output, this goes for any input at all. Add 'hellfire tony pony' in there why the hell not:  KakerMix fucked around with this message at 15:17 on May 31, 2023 |
#
?
May 31, 2023 14:52
|
|
|
|
| # ? May 30, 2024 20:59 |
|
|
Shakespeare in the pork   
|
|
#
?
May 31, 2023 15:02
|
|
|
made some backdrops for my kids' school portraits, screw you "we have lasers!" 60s candy mascots pattern flat wall blue  african pattern wallpaper lithograph  bauhaus abstract geometric wallpaper  colorful amplifiers pattern wall flat straight lithograph  colorful boomboxes pattern wall flat straight lithograph  colorful mid century soviet cyrillic matchbook covers pattern straight  colorful mid century soviet cyrillic matchbook covers pattern straight  colorful patterned skateboard decks wallpaper  colorful patterned skateboard decks wallpaper  colorful speakers lithograph wallpaper flat straight  complex flat chess pieces duotone lithograph wallpaper  complex origami flat blueprint schematics wallpaper  complex origami flat blueprint schematics wallpaper  complex origami flat blueprint schematics wallpaper  flat chess pieces charts lithograph wallpaper  flat chess pieces charts lithograph wallpaper  flat chess pieces charts lithograph wallpaper  hand painted speakers lithograph wallpaper flat straight  mid century minimalist eames wallpaper pattern  patterned skateboard decks wallpaper lithograph  patterned skateboard decks wallpaper  pop art speakers lithograph wallpaper flat straight  retro ice cream confections pattern wallpaper 
|
|
#
?
May 31, 2023 15:14
|
|
|
    "I sleep in a racing car, do you?" Screenshot from a late 90s infomercial selling a racecar bed, dad with a mullet
|
|
#
?
May 31, 2023 15:21
|
|
|
KakerMix posted:'poolcore' This needs to be the theme of the day.
|
|
#
?
May 31, 2023 15:36
|
|
|
SD latches onto the word pool in 'poolcore' pretty hard and thinks you just want expensive looking indoor pools.
|
|
#
?
May 31, 2023 15:50
|
|
|
Sedgr posted:SD latches onto the word pool in 'poolcore' pretty hard and thinks you just want expensive looking indoor pools. I very much do, thank you.
|
|
#
?
May 31, 2023 15:51
|
|
|
KakerMix posted:
hell yeah
|
|
#
?
May 31, 2023 16:03
|
|
|
Sedgr posted:SD latches onto the word pool in 'poolcore' pretty hard and thinks you just want expensive looking indoor pools. I am super down for poolcore, since it seems a lot like those aquarium houses I was making and I would love to make more 
|
|
#
?
May 31, 2023 16:04
|
|
|
Seems like a theme to me! These places look expensive.   
|
|
#
?
May 31, 2023 17:01
|
|
|
Holy poo poo yes. I tried a prompt like this with Bing (monster trucks in the colliseum, gladitorial combat, surrounded by Roman crowds, with Roman cheerleaders visible), but the results weren't anywhere near as convincing.
|
|
#
?
May 31, 2023 17:09
|
|
|
XYZAB posted:Holy poo poo yes. I tried a prompt like this with Bing (monster trucks in the colliseum, gladitorial combat, surrounded by Roman crowds, with Roman cheerleaders visible), but the results weren't anywhere near as convincing. Different models have different strengths. Even with bing though keep trying and rewording. You can usually find something even if you have to get creative. You can make planets in most things with "fantasy hand drawn map wrapped around a sphere floating in space" generic planet will mostly just get Earth clones. Do not be afraid to get creative and just describe what feature you want to represent something to get the image you want.
|
|
#
?
May 31, 2023 17:13
|
|
|
Here, enjoy some Synthwave Poolcore Cyberpunk  
|
|
#
?
May 31, 2023 17:15
|
|
|
pixaal posted:Different models have different strengths. Even with bing though keep trying and rewording. You can usually find something even if you have to get creative. I've just about got Stable Diffusion up and running at home, but the choices to make between models is what I'm stuck at. Is there a good/better/best one that everyone is more or less agreed upon? ymgve posted:possible PTSD from watching Norwegian 80s children shows?  What was the deal with old kids' shows and giraffes coming in through the windows?
|
|
#
?
May 31, 2023 17:18
|
|
|
 Am I doing this 'poolcore' thing wrong?
|
|
#
?
May 31, 2023 17:21
|
|
|
water for everyone!     edit: too wet, let's dry you off   "When you only see one pair of footprints, that was when I carried your stack of corporate binders" ymgve fucked around with this message at 17:36 on May 31, 2023 |
|
#
?
May 31, 2023 17:27
|
|
|
pixaal posted:Different models have different strengths. Even with bing though keep trying and rewording. You can usually find something even if you have to get creative. I'm wondering if Bing handles language differently, or if it makes passable images for almost any prompt because the parameters are set for "general purpose." Thanks y'all for the advice.
|
|
#
?
May 31, 2023 17:36
|
|
|
XYZAB posted:Holy poo poo yes. I tried a prompt like this with Bing (monster trucks in the colliseum, gladitorial combat, surrounded by Roman crowds, with Roman cheerleaders visible), but the results weren't anywhere near as convincing. The real trick is to ask for the hot tub seats. Use code POOLCORE to book yours today! 
|
|
#
?
May 31, 2023 17:44
|
|
|
Can you even call yourself an angel investor tech startup if you don't have a kooky office space?
|
|
#
?
May 31, 2023 17:47
|
|
|
             
|
|
#
?
May 31, 2023 17:47
|
|
|
put me in the back rooms equivalent of a lazy river and forget me cause I'm not getting out.
|
|
#
?
May 31, 2023 17:52
|
|
|
XYZAB posted:I've just about got Stable Diffusion up and running at home, but the choices to make between models is what I'm stuck at. Is there a good/better/best one that everyone is more or less agreed upon? If you want just Stable Diffusion models, I was using it even more generally to include MJ Bing and Dalle. Honestly I used 1.5 and 2.1 for a very long time like until March I was just using them with control net and other plugins, get to know the base model and how it's worded most other models are just trained on to of 1.5, so learning to make good images with base 1.5 is a good starting point. I currently like normal Civitai warning  content exists here and may exist in comment images for any model or lora. https://civitai.com/models/10540/verisimilitude verisimilitude for realistic https://civitai.com/models/9409/or-anything-v5 anything v5 for anime and I really haven't found a good general art one I usually just go for a prompt that's vague and prompting styles then run it against like 5-6 models then refine on whichever model I feel had the most interesting set of 16 images or whatever I generated to test. Be sure to grab .safetensor versions .pickle can in theory have executable code and could house a trojan. I don't think any actual examples have been given of one in the wild but why not just wait for the safe one to get uploaded if they don't someone else will likely do it and post a link in the comments if the model is any good. pixaal fucked around with this message at 18:03 on May 31, 2023 |
|
#
?
May 31, 2023 17:54
|
|
|
KakerMix posted:I have no idea what 'poolcore' is but when you ask MidJourney what a picture is with the /describe function it ends up suggesting a lot of ----core words so it must mean something to it. xcore works much like xpunk does, should we try to find a bunch of xwords to share?  poolcore Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 1, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Version: v1.2.1  poolcore Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 4, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Denoising strength: 0.48, Version: v1.2.1, Hires upscale: 1.5, Hires steps: 32, Hires upscaler: 4x-UltraSharp  poolpunk Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 1, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Version: v1.2.1  poolpunk Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 4, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Denoising strength: 0.48, Version: v1.2.1, Hires upscale: 1.5, Hires steps: 32, Hires upscaler: 4x-UltraSharp XYZAB posted:
These are some older models I can personally vouch for. I know you can get quality out of these. Protogen Infinity seek.art MEGA and related seek.art MEGA Midjourneyv4 DreamLike mix Analog Diffusion. However this model has been stripped down to just a LoRA, is 1.7 gigs smaller, and can now be used with a different model. If you're going for specific camera settings it's a good start at least. These are newer generations and the jump in quality is noticeable. Rev Animated Realistic Vision I haven't tried these latest generations yet but they look promising. M4rv3ls and Dungeons Xenoengine Cyberrealistic If you click the 'Script' dropdown at the bottom you'll find "X/Y/Z Plot". This allows you to make grids of images so you can compare settings and models to see for yourself what each thing does. Grids are your friend. On settings like 'Samplers' there is a yellow box you can click to auto fill all available so you can compare them all at once. Important note that 'Hires. fix' upscaling does a tremendous amount of work in fixing up and adding details on an image. It's definitely something you want to learn how to use. And finally for MUCH further reading, https://sdtools.org/. This giant wheel has doubled in size since I first linked it and I'm personally not even half-way through some of these. Still haven't even started using ControlNet lol. I say it's ok to feel lost in all this, just have fun with it and see what you can come up with. Be sure to share with the class! These are excellent monster trucks!
|
|
#
?
May 31, 2023 18:13
|
|
|
I prefer to lock a seed and add remove words switch models and learn that way, grids are a good comparison tool but I couldn't really make good use of them until I had a solid feel for SD. Now it's just automatic if I want something dripping I know if moist misted or wet is the better choice automatically, not wet enough? add more! Having one hell of a time right now to get a poolcore kitten laser pointer fight but I will find the combination of words that doesn't give me melted cat face! *sigh* okay adding human to the negative stop giving me naked women in this pool! Why does the pool make this model so horny! holy poo poo that was it human negative instantly good cats! Now to get the composition how I want.  e: okay nope we're done even with a stack of negatives getting it more towards a photo reads topless woman sunbathing in pretty much everything that isn't the base model. I'm thinking it's pool and kitten in the same prompt. base 2.1 might work but I think I'll try something else with pools pixaal fucked around with this message at 18:59 on May 31, 2023 |
|
#
?
May 31, 2023 18:17
|
|
|
I've found the SD checkpoints Deliberate, Realistic Vision, RevAnimated, and Dreamshaper all to be different enough to be useful.
|
|
#
?
May 31, 2023 18:37
|
|
|
 
|
|
#
?
May 31, 2023 18:48
|
|
|
giraffe on steroids djinn / genie giraffes   imagine four balls on the edge of a cliff... AI works the same way     Soulhunter fucked around with this message at 19:01 on May 31, 2023 |
|
#
?
May 31, 2023 18:56
|
|
|
|
|
#
?
May 31, 2023 19:39
|
|
|
is there an AI bot that combines two songs, ala Girl Talk
|
|
#
?
May 31, 2023 19:51
|
|
|
brocked posted:made some backdrops for my kids' school portraits, screw you "we have lasers!" These are awesome, too bad they don’t tile or I’d have an endless wallpaper rotation.
|
|
#
?
May 31, 2023 20:31
|
|
|
Dicty Bojangles posted:These are awesome, too bad they don’t tile or I’d have an endless wallpaper rotation. Automatic 1111 the most popular frontend for Stable Diffusion supports tiles out of the gate - make whatever you want, or maybe we can do a tile day I don't think many people have really posted any tiled images I think I've seen 1 or 2 people other than myself ever post one. 
|
|
#
?
May 31, 2023 20:34
|
|
|
Tiling is kind of hit or miss, sometimes its tough to make it follow much of a prompt.      
|
|
#
?
May 31, 2023 21:03
|
|
|
Ramrod Hotshot posted:is there an AI bot that combines two songs, ala Girl Talk https://rave.dj/ did that, haven't used it in a while but should still be operational
|
|
#
?
May 31, 2023 21:04
|
|
|
no matter how much I try to keep the negatives reasonable after 3-4 loops of upscaling and downscaling I end up with either half naked women sunbathing by the pool or people that look more like Microsoft Emojis Honestly I think I like the base image better bah (It has no idea what a log flume is other than found in a theme park) Looking at it again I guess the original does have some flesh mounds. log flume, poolcore, disco rave, sharp focus, exciting, 4k <lora:steampunkai:0.33> Negative prompt: captcha, font, text, boring, dull, sleepy Steps: 20, Sampler: DDIM, CFG scale: 13, Seed: 389273557, Size: 768x512, Model hash: 79939acf90, Model: verisimilitude_v2, Version: v1.2.1  
|
|
#
?
May 31, 2023 21:19
|
|
|
pixaal posted:Honestly I used 1.5 and 2.1 for a very long time like until March I was just using them with control net and other plugins, get to know the base model and how it's worded most other models are just trained on to of 1.5, so learning to make good images with base 1.5 is a good starting point. KwegiboHB posted:I'm not sure if there's a widely accepted 'general model' so here are a bunch. Hell yeah, thanks to you both. Is there a way I can train a model off of the contents of an 80gb folder of images I've been collecting for the last 15 years? It's mostly old SA memes/photoshop phridays/vacation photos, but I think it'd be interesting to use a dataset that I'm already intimately familiar with just to see how SD interprets and remixes it. Edit: Google says yes, yes I can, but if anyone has a link or a hot tip they wanna throw at me to make it easier, I'm all for it.
|
|
#
?
May 31, 2023 21:46
|
|
|
https://twitter.com/hausofdecline/status/1664006436463026176?s=21
|
|
#
?
May 31, 2023 22:18
|
|
|
Computer, show me goon caves
|
|
#
?
May 31, 2023 22:20
|
|
|
Sumerian Telecom posted:Is AI art good now? Has it gotten good yet? I haven't checked in a while but am very curious - what do you think? Why ask us? Ask the source… MJ Prompt: “Is AI art good now?”  I think that means yes.
|
|
#
?
May 31, 2023 23:13
|
|
|
Doctor Zero posted:Why ask us? Ask the source… Is this based on Data's daughter's appearance? I'm not sure why it reminds me of her directly comparing they look different but something keeps nagging.
|
|
#
?
Jun 1, 2023 00:37
|
|
|
|
| # ? May 30, 2024 20:59 |
|
|
Looks a little like there's some Krysten Ritter DNA in the mix there too, to me. e: Prompting for Krysten Ritter weaves in some Kate Micucci and Zooey Deschanel. We're approaching manic pixie dream girl critical mass. Krysten Ritter, hair in the style of bangs, red shirt 
Soulhunter fucked around with this message at 00:59 on Jun 1, 2023 |
|
#
?
Jun 1, 2023 00:53
|
|