|
any good youtube follow along type tutorials? Y'all have been excellent help so far.
|
#
?
Apr 2, 2024 16:40
|
|
|
|
| # ? May 14, 2024 04:29 |
|
|
loving around with using screen caps of old edutainment games as style references
|
|
#
?
Apr 2, 2024 19:45
|
|
|
Since when does this "select" button exist in GPT-Dalle? Pretty good inpainting "The cat has glowing eyes" "the cat is sitting on a grassy hill"   
|
|
#
?
Apr 2, 2024 21:02
|
|
|
Just started yesterday as far as I know. It works fairly well.
|
|
#
?
Apr 2, 2024 21:02
|
|
|
            
|
|
#
?
Apr 3, 2024 19:01
|
|
|
Just installed Stability Matrix and holy poo poo. 
XYZAB fucked around with this message at 04:41 on Apr 6, 2024 |
|
#
?
Apr 6, 2024 03:03
|
|
|
Oh neat, it's very high-res what model is that? I've been playing with OpenPose, everything here uses this same pose as input:            
|
|
#
?
Apr 6, 2024 05:10
|
|
|
 
AARD VARKMAN fucked around with this message at 06:46 on Apr 6, 2024 |
|
#
?
Apr 6, 2024 06:30
|
|
|
https://i.imgur.com/Cp0yM8T.mp4 Eat it Sora.
|
|
#
?
Apr 6, 2024 07:16
|
|
|
             
|
|
#
?
Apr 6, 2024 17:42
|
|
|
gooncave.jpg
|
|
#
?
Apr 6, 2024 19:10
|
|
|
snorch posted:Oh neat, it's very high-res what model is that? With those two I was basically just loving around in Stability Matrix's in-house "Inference" package interface (which has a ComfyUI backend), using the AfroditeXL porn model as a base with EpicRealism XL as a refiner. I downloaded a bunch of random models from the built-in model browser and have been trying to comprehend how models and refiners work together, and why some models won't work as a model and some won't work as refiners. It's all incredibly confusing to me right now. As for the high-res comment, I was then experimenting with HiResFix and Upscaler steps, but the end result might be more from having a 16gb GPU than from any of the models I used. I had previously used an 8gb 2070 Super and a 12gb 3080 Ti, and both of them had problems upscaling even the most modest requests, telling me that my requests would need around 14gb to complete. So I got a 16gb 4070 Ti Super and I've not encountered this problem since, and I've noticed that it's been using pretty consistently around 14gb while rendering. Prior to installing Stability Matrix, I had an intuitive comprehension of how these procedures should all play off of each other down the line from one to the next, as we all might from sort of just gleaning concepts from one another as this thread in particular has sort of evolved, but seeing these images be created out of thin air, step-by-step, has me questioning how individuated each process is, and how wrong my intuition has been. Or maybe individuated isn't the right word. When I tell AfroditeXL to generate an image in 10 steps, it generates 10 steps of what that model was trained to generate, which is more or less porn. But, if I tag on a refiner, in this case the basic EpicRealism model, it seems that the image from the final step of stage one, which is more or less porn, is then taken as the base image for an entirely new round of steps in the EpicRealism model, which basically removes all of the attributes that the Afrodite model gave it in the first place, minus maybe the broadest implication of form provided by the previous model at the moment that the new model took over. Intuitively, I sort of expected them to mix and have features of both. What I'm getting in reality is more like an erasure of the first model's work, which is completely overwritten by the second (refiner) model's work. And then, to make matters more confusing, HiresFix is supplied a user-defined number of steps to make your image larger by an also-user-defined multiplication factor, but instead of simply keeping the image looking mostly how it came out at the end of the refiner stage, it garbles it up even more and starts adding features that are something like (1/x) the size of any of the most noticeable patterns of the original end-result, where x=the multiplication factor you gave it. So although you're increasing the resolution of the image that was finalized at the end of the refiner step, the hires process then starts blowing things up in both a literal and figurative sense. Image is larger, but is now also comprised of a bunch of smaller patterns that used to comprise the broader image when it was smaller. This is especially noticeable in images with skulls. If you had a relatively normal looking skull at the end of the refiner step, the hires step takes that and starts adding another line of teeth where it sees a horizontal line, or it starts adding eye sockets in random places where it notices circular visual patterns, and the end result looks like a mess. I see this in a lot of the images generated by guys who have massively overpowered systems with A100s or Quadro 8000s with 50+GB of VRAM. It's cool that they have that much power available, but they seem to be completely misunderstanding that the results they're getting (that look like cascades of self-similar image properties) are just hallucinations brought about by having way too much power. Or maybe this just means SD isn't scaled properly. I literally don't know how any of this works in the backend, I can only comment on what I see based on observation and using the tools with what I have available to me. AI watching me try to put into words how images are crafted by the tools it gave me and that I don't understand, must be like how grass feels when it contemplates how it successfully tricked humans into evolving it into wheat and grain.
|
|
#
?
Apr 6, 2024 19:41
|
|
|
I'm STILL giggling about these images and my wife is upset that I am so easily entertained.
|
|
#
?
Apr 6, 2024 19:51
|
|
|
Works on contingency? No, money down!
|
|
#
?
Apr 6, 2024 21:09
|
|
|
|
|
#
?
Apr 6, 2024 22:27
|
|
|
  Francis Bacon / Salvador Dali crossover.
|
|
#
?
Apr 7, 2024 01:28
|
|
|
XYZAB posted:stuff Yeah to me it's a wide mysterious world but all the youtube tutorials have been super helpful in wrapping my head around how people actually work with these things. Some more experiments today, this time it's mostly running JuggernautXL in two stages, through two KSamplers that have been fed separate prompts. It's good for doing variations on an overarching theme, you can freeze the seed on the first stage and just run 10 steps or so of the finisher stage. Simple prompts seem to still work best, I've gotten so much mileage out of the primer "party jesus" alone: weed party jesus/budgie smugglers  dog/cat  skyrim/photography  Bigger gallery here. This is spooky and wet and moody and I love it. snorch fucked around with this message at 02:38 on Apr 7, 2024 |
|
#
?
Apr 7, 2024 02:22
|
|
|
Sleeping with the fishes?
|
|
#
?
Apr 7, 2024 16:26
|
|
|
How being bored on a Sunday turned into Slimepocalypse. It started off innocently enough: after going to a coffee shop, I idly wondered if Bing could create the same kind of Italian stock photos the place used to decorate its walls.  It could, and they looked fine, but a bit bland. So then I thought: what if I add an alien?  The alien was rather conventional and man-in-a-suit-y. How about something more squishy and organic?   Kinda cute, but that wasn't what I wanted. These aliens should be a bit more... absorbent.     And maybe terrifying. Or just plain gross.      Why not expand beyond a small Italian town square?     Maybe go full Thing while we're about it.     I have no idea why Nicolas Cage turned up when things moved indoors, but here he is.  And now the descent into full Blob insanity begins. Thanks, Bing!      Not even the security of booze can protect you from the meldown madness.    Inevitably, the slime takes Manhattan. Don't let it touch you!   
|
|
#
?
Apr 8, 2024 10:48
|
|
|
A fight between personifications of reddit.com and somethingawful.com
|
|
#
?
Apr 8, 2024 12:31
|
|
|
axolotl farmer posted:A fight between personifications of reddit.com and somethingawful.com I like the 3rd hand coming out of the goons right leg.
|
|
#
?
Apr 8, 2024 14:03
|
|
|
^I tried a few variations of that prompt and it just gives both combatants a Snu head 
|
|
#
?
Apr 8, 2024 14:14
|
|
|
Sab669 posted:^I tried a few variations of that prompt and it just gives both combatants a Snu head Took a bit but I got bing with a brawl between personifications of Reddit and Somethingawful.com they should be grappling.  GPT had some interesting ideas when I gave it the concept but it is confusing the two characters  e: I made some pictures of a moron looking at the sun during an eclipse    pixaal fucked around with this message at 19:30 on Apr 8, 2024 |
|
#
?
Apr 8, 2024 14:25
|
|
|
Just for fun, I entered the prompt "Monty Python" into SD with no other parameters at all...   
|
|
#
?
Apr 9, 2024 01:03
|
|
|
Louis Wain twitches in his grave.
|
|
#
?
Apr 9, 2024 02:22
|
|
|
            
|
|
#
?
Apr 9, 2024 23:38
|
|
|
https://www.udio.com Suno challenger by some ex-Google/Deepmind people. https://twitter.com/udiomusic/status/1778045322654003448 https://twitter.com/Jakestein776/status/1777912745892966498 https://twitter.com/Jakestein776/status/1777913344516604273 https://twitter.com/DotCSV/status/1777982585026531816 https://twitter.com/lifeafterAi_/status/1776930684642443400 BoldFace fucked around with this message at 14:59 on Apr 10, 2024 |
|
#
?
Apr 10, 2024 14:51
|
|
|
oh you have it at the top of the post how did I miss that!
|
|
#
?
Apr 10, 2024 15:08
|
|
|
pixaal posted:
My bad, I forgot to add it to my original post, so I shamelessly stealth edited it.
|
|
#
?
Apr 10, 2024 15:17
|
|
|
That site just absolutely ate poo poo, lol Only got to play with it a bit but it does classical really well.
|
|
#
?
Apr 10, 2024 16:04
|
|
|
As a musician, can I just say, LOL. Nashville's never hiring anybody real ever again. I've been messing with a new workflow, using IP-Adapter with a fixed image in the composition input, swapping in random images in the style input, and providing a simple prompt like "biker keg party". I'm really in love with the control it gives over the output, it's like a lora you don't have to train. For today's dump, some poster art.               
snorch fucked around with this message at 23:56 on Apr 10, 2024 |
|
#
?
Apr 10, 2024 23:01
|
|
|
Udio servers are tanked but I'm looking forward to playing with it more when it's stable. Couple gens: Spaghetti Western Xeno vs Ripley https://www.udio.com/songs/nLWtown3Q4m7mHqA98fWYY Kong vs Godzilla rap (seems to skip a word here and there) https://www.udio.com/songs/oy2ZnbNeMw6PZawdsHsJQJ Hair metal - BABA HAS GUN (PUTTY BETTER RUN) https://www.udio.com/songs/7B3AwVfGqYJXrFXsEPpRjA spacetoaster posted:I'm STILL giggling about these images and my wife is upset that I am so easily entertained. 1-800-POSSLAW https://www.udio.com/songs/5F92XUsWfsgvCzV8mKPw6D Bonus POSSLAW in Suno https://suno.com/song/e6455ce1-6b70-4f75-a701-28c95e248c4f
|
|
#
?
Apr 11, 2024 00:21
|
|
|
|
|
#
?
Apr 11, 2024 02:35
|
|
|
Funny, of all things I've never been able to get SD to paint a skateboard that isn't misshapen.
|
|
#
?
Apr 11, 2024 03:15
|
|
|
So uh I make "art" I guess: https://twitter.com/AngstHarrier/status/1778304981587886136
|
|
#
?
Apr 11, 2024 07:20
|
|
|
snorch posted:As a musician, can I just say, LOL. same. i havent really been impressed with music ai stuff (other than certain noise reduction tools) and ive been hesitant to mess around with it since its my actual job but its definitely catching up The Tragedy of Darth Plagueis The Wise
|
|
#
?
Apr 11, 2024 21:54
|
|
|
I've already got 3 Tracks done for this Psych-CyberOpera. https://www.udio.com/songs/3KzFdae3iaqYJNGeoLCMxW This is definitely interesting from a production standpoint. The lyrics and how it puts in background vocals is interesting. I've just been listening to the staff picks on my studio monitors and what its outputting sounds for the most part is not awful, like better then local produced stuff. What if HP Lovecraft was Thom Yorke and Had Greenwood as a guitarist? https://www.udio.com/songs/t9KtDPeRCLR3Li9kSJ5vpQ
|
|
#
?
Apr 12, 2024 09:45
|
|
the Mods, they knew!
|
TheGoonspiracist posted:I've already got 3 Tracks done for this Psych-CyberOpera. this one sounds good, lotta AI songs sound muddy.
|
|
#
?
Apr 12, 2024 18:08
|
|
|
https://voca.ro/1kZBR1OJU2ws Udio too can do stand up comedy stuff, and it's *better* at it than Suno. 33 seconds at a time makes it different than Suno, but the depth to the generations is pretty striking. You can hear it in a lot of things that Udio outputs, but stand up and crowd cheering make it quite clear. The way the crowd laughs, noises, hear little cheers and stuff in the background. Like a perfect illustration on what upping the quality does.
|
|
#
?
Apr 12, 2024 18:09
|
|
|
|
| # ? May 14, 2024 04:29 |
|
|
KakerMix posted:https://voca.ro/1kZBR1OJU2ws That's incredible. My wife: "it's not really that funny" Me:  a year and a half ago a static image of an astronaut riding a unicorn was front page news, but "not very funny" is your entire reaction to this?
|
|
#
?
Apr 12, 2024 21:33
|
|