|
BrainDance posted:Quick question for anyone who's used the basujindal optimized fork of SD and automatic1111's webui. I have both, and I'm messing with the webui now. But, for img2img I cant seem to find an option that works similarly to the --strength option in the basujindal fork. Does that not exist? Or am I blind? I can't even find much documentation on what --strength actually changes, but it was probably the most important option for img2img because, without it, I'm getting cool stuff but it looks nothing like the source image. "Denoising strength" is probably what you're looking for.
|
# ? Oct 1, 2022 14:12 |
|
|
# ? May 30, 2024 09:47 |
|
Ima Computer posted:"Denoising strength" is probably what you're looking for. Thanks, yeah that makes sense since it's right there, and turning it down does do what adjusting --strength would do. Didn't consider it because it really doesn't seem like "denoising" is the right word for what happens when it's changed. I mean I guess it is, it just didn't seem like it would be.
|
# ? Oct 1, 2022 14:18 |
|
The guy who makes Diffusion Bee is also working on this, apparently: https://liner.ai/ "a free tool which lets you build and deploy machine learning applications within minutes. No coding or expertise in machine learning required." So in theory (have not tested it) you can use it to train the AI with your characters/photos/concepts, with a nice UI, and export them to use in other AI apps (objective being of course to make art with your own ideas). Like the programs in recent pages, just with a nice UI. This one works with both Windows and Mac Comfy Fleece Sweater posted:I posted this earlier - For Apple M1 Mac owners:
|
# ? Oct 1, 2022 14:36 |
|
Wheany posted:You've probably seen people posting goatse and tubgirl that have been run though img2img, and I did some very quick testing and it seems like the trick is using a high CFG scale and a pretty low denoising strength. Thanks, I've been wondering how those were made but didn't have time to gently caress around with img2img too much. Depends on the source image I guess but for this one I had to bring down denoising to about 0.35 to get recognizable results.
|
# ? Oct 1, 2022 14:47 |
|
Cable Guy posted:What engine is this? Stable Diffusion...? Are you using it with a local install? Yes, a local install of AUTOMATIC1111 stable diffusion webui, in the img2img tab, using whatever the official model is called, not "waifu diffusion" Wheany fucked around with this message at 15:30 on Oct 1, 2022 |
# ? Oct 1, 2022 15:28 |
|
mobby_6kl posted:Thanks, I've been wondering how those were made but didn't have time to gently caress around with img2img too much. Depends on the source image I guess but for this one I had to bring down denoising to about 0.35 to get recognizable results. Lol she seems a little indignant about art cliches.
|
# ? Oct 1, 2022 15:47 |
|
Tyger Tyger burning bright In the forests of the night, What immortal hand or eye Could frame thy fearful symmetry Burnt in distant deeps or skies The cruel fire of thine eyes Could heart descend or wings aspire What the hand dare seize the fire And what shoulder and what art Could twist the sinews of thy heart And when thy heart began to beat What dread hand and what dread feet When the stars threw down their spears And water'd heaven with their tears Did he smile his work to see Did he who made the lamb make thee - Tyger Tyger burning bright In the forests of the night, What immortal hand or eye Could frame thy fearful symmetry William Blake would have loved and hated this tech. Humbug Scoolbus fucked around with this message at 16:02 on Oct 1, 2022 |
# ? Oct 1, 2022 15:56 |
|
/imagine "yourself"
|
# ? Oct 1, 2022 16:07 |
|
A goon just posted about buying this in PYF so I thought... why shouldn't I have one as wellwidefault posted:And another piece for the music room wall a rough pencil sketch of a jimi hendrix playing electric guitar, by George Manson (the name was from the interrogation)
|
# ? Oct 1, 2022 16:49 |
|
"the animals of hell painted by Albrecht Dürer painted by Salvador Dali painted by Francis Bacon painted by Philip Guston painted by Rachel Ruysch painted by Picasso"
|
# ? Oct 1, 2022 18:56 |
|
|
# ? Oct 1, 2022 20:42 |
|
I'm trying to get StableDiffusion on my machine and I think I'm a bit out of my depth. I followed the guides/troubleshooting as close as I could and can get to here: At this point it just sits here, I can't type in this window, and the guides all just kinda end, so I'm at a loss. Help?
|
# ? Oct 1, 2022 21:42 |
|
Evilreaver posted:I'm trying to get StableDiffusion on my machine and I think I'm a bit out of my depth. I followed the guides/troubleshooting as close as I could and can get to here:
|
# ? Oct 1, 2022 21:43 |
|
Oh thank god, I had a feeling I had face-planted at the finish line. Now all is well. Many thanks!
|
# ? Oct 1, 2022 21:57 |
|
A few more attempts though this one seems a bit more difficult to nail down
|
# ? Oct 1, 2022 22:01 |
|
The Waifu Diffusion people are currently now training the model on 660K images from danbooru (up from just 56K images in wd-1.2) with 10 epochs planned. They're up to 5 epochs on this dataset and have released some checkpoints based on that: https://huggingface.co/hakurei/waifu-diffusion-v1-3/tree/main
|
# ? Oct 2, 2022 00:23 |
|
Grabbed this prompt from Reddit, but it got good results. A magazine ad for a breakfast cereal called "Blood and Teeth." This is the kind of creepy that I love. "Declaration of Independence" Why do they all have the same face in each pictures? That's kind of freaky. Vlaphor fucked around with this message at 02:24 on Oct 2, 2022 |
# ? Oct 2, 2022 01:52 |
|
holy poo poo wow
|
# ? Oct 2, 2022 04:54 |
|
I am tempted to print this and hang it on my wall linked because nsfw if thumbnailed, or on the phone app The goats are
|
# ? Oct 2, 2022 07:31 |
|
Vlaphor posted:Why do they all have the same face in each pictures? That's kind of freaky. Founding fathers meant "singular father, plural" Also loving the dudes peeking from behind the wall drape and under the chair
|
# ? Oct 2, 2022 07:44 |
|
mobby_6kl posted:Thanks, I've been wondering how those were made but didn't have time to gently caress around with img2img too much. Depends on the source image I guess but for this one I had to bring down denoising to about 0.35 to get recognizable results. Good loving lord,
|
# ? Oct 2, 2022 08:26 |
|
Just learned about Loab. https://www.youtube.com/watch?v=i9InAbpM7mU What I find most infuriating, is that there a now like a million articles about this, and NONE of say what AI this was created on. Like that's pretty key, isn't it? Anyway, I'm assuming Stable Diffusion (probably).
|
# ? Oct 2, 2022 08:50 |
|
Vlaphor posted:Why do they all have the same face in each pictures? That's kind of freaky. The dataset for the founding fathers is too similar, and doesn't differentiate who each founding father is. The token gets assigned to a single character, and so all of them are merged together to make that guy.
|
# ? Oct 2, 2022 09:01 |
|
Wheany posted:Yes, a local install of AUTOMATIC1111 stable diffusion webui, in the img2img tab, using whatever the official model is called, not "waifu diffusion"
|
# ? Oct 2, 2022 09:12 |
|
Humbug Scoolbus posted:
I don't even care what the critics say. This is sublime. These digital things... computational machines... loaded with a scarce few gigabytes of numeric network weights are somehow evoking imagery that represents a distilled condensate of us, our art, our sensibilities. Yes, these are still guided by human hand and discerning curation, clever prompting, reverse engineered spells. So much chaff to discard. But what comes out is surely art.
|
# ? Oct 2, 2022 10:39 |
Somebody made a really good point in another thread; If these AIs are producing art (whatever that means) a clear test would be whether an AI trained only on previous AI images can produce at the same level as the original model, or whether "copy of a copy" type degradation/hominization sets in. We may see this play out regardless, as AI generated content swamps human-only content and becomes the majority of a naïve or auto-updating training set. Another angle I've been thinking about is AI text - I've seen coherent paragraphs-long snippets generated from just a few words of prompt and maybe a starter line. But would you read an AI short story (or novel)? I'm sure you could get 5,000 words that follow some characters through a comprehensible plot. But my instinct is that all the stuff that happens on the level other than "sentence by sentence" would be very lacking. Grammar is not prose, plot is not story. Coherence is not style, comprehensibility is not theme. Named or recognizable entities are not character. Legibility is not resonance. And so on. Not to say all (or most!) human authors are excelling in all these "beyond the wiki summary" levels of the art, but the good ones are hitting most of them most of the time. I don't know if I'll ever have much desire to read an AI novel, or watch an AI film. Then again, how different would an AI film or show be from the over-produced, executive/studio-meddled, algorithm-guided bland slop we already get?
|
|
# ? Oct 2, 2022 16:05 |
|
I've been thinking and I have One Good Question for yall. How does this stuff present an advantage to using other forms of computer or machine art? It seems to me 3d modeling, photoshop and so on have the advantage of offering very precise control for stuff that is hard to control in physical processes. What is the advantage of using this tool over those, when the tool basically has to guess what you are thinking of before you land on it?
|
# ? Oct 2, 2022 16:20 |
|
No, you haven't been thinking.
|
# ? Oct 2, 2022 16:43 |
|
FunkyAl posted:I've been thinking and I have One Good Question for yall. How does this stuff present an advantage to using other forms of computer or machine art? It seems to me 3d modeling, photoshop and so on have the advantage of offering very precise control for stuff that is hard to control in physical processes. What is the advantage of using this tool over those, when the tool basically has to guess what you are thinking of before you land on it? 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. Another great usage, which is in large part due to the speed of which the algorithm operates, is sketching. Say I have an idea in my head for an art-piece, a fight scene perhaps, showing off a cool character. But I'm not fully locked into what exactly the background should be. I type in "My Cool Character Description" and then a few different arena or background words, and now within seconds I can see what fits the action sequence best. Or maybe I do know both what the character is like, what they are doing and where they are doing it, but I cant' figure out what pose is best. I could spend an hour fiddling around in blender, fifteen minutes in Dreams (for the ps4), or a lifetime in photoshop, or I could tell the AI to just make fifteen variations of the same prompt.
|
# ? Oct 2, 2022 16:43 |
|
FunkyAl posted:I've been thinking and I have One Good Question for yall. How does this stuff present an advantage to using other forms of computer or machine art? It seems to me 3d modeling, photoshop and so on have the advantage of offering very precise control for stuff that is hard to control in physical processes. What is the advantage of using this tool over those, when the tool basically has to guess what you are thinking of before you land on it? Amazing? Not really, but how else could I do this in 2 minute?
I didn't really have a reason to make this image, but if I wanted a custom invitation card for my halloween party, that's a great way to get it done quickly.
|
# ? Oct 2, 2022 16:46 |
|
Prolonged Panorama posted:Somebody made a really good point in another thread; If these AIs are producing art (whatever that means) a clear test would be whether an AI trained only on previous AI images can produce at the same level as the original model, or whether "copy of a copy" type degradation/hominization sets in. We may see this play out regardless, as AI generated content swamps human-only content and becomes the majority of a naïve or auto-updating training set. AI text simply isn't there yet. It will probably take a long time for that to happen, since it doesn't actually have general knowledge. It can be used as writing prompt for ideas. Training AI on AI or procedurally generated data is actually a thing that is done already, and can be quite useful. It can be hard to tag stuff, hard to get a large enough dataset, and you don't have to worry about extra unwanted things. A lot of datasets are already hand curated.
|
# ? Oct 2, 2022 16:51 |
|
FunkyAl posted:I've been thinking and I have One Good Question for yall. How does this stuff present an advantage to using other forms of computer or machine art? It seems to me 3d modeling, photoshop and so on have the advantage of offering very precise control for stuff that is hard to control in physical processes. What is the advantage of using this tool over those, when the tool basically has to guess what you are thinking of before you land on it? Time and cost. That's the kind of advantages capitalism loves. Obviously the technology isn't there yet to fully replace human artists, but the rate of progress in the last years is accelerating so who knows what these tools will be able to produce in a few years? I'm not saying I think this is desirable. Just that what computers can create today is nothing short of amazing.
|
# ? Oct 2, 2022 16:53 |
|
Prolonged Panorama posted:Somebody made a really good point in another thread; If these AIs are producing art (whatever that means) a clear test would be whether an AI trained only on previous AI images can produce at the same level as the original model, or whether "copy of a copy" type degradation/hominization sets in. We may see this play out regardless, as AI generated content swamps human-only content and becomes the majority of a naïve or auto-updating training set. It'd degrade. They're just using probability to guess where pixels should go. If you train something designed to approximate based on nothing but approximation then it's going to start skewing.
|
# ? Oct 2, 2022 16:56 |
|
Turning and turning in the widening gyre The falcon cannot hear the falconer Things fall apart; the centre cannot hold; Mere anarchy is loosed upon the world The blood-dimmed tide is loosed, and everywhere The ceremony of innocence is drowned The best lack all conviction, while the worst Are full of passionate intensity. -- Surely some revelation is at hand; Surely the Second Coming is at hand. Troubles my sight: somewhere in sands of the desert A shape with lion body and the head of a man A gaze blank and pitiless as the sun, Is moving its slow thighs, while all about it Reel shadows of the indignant desert birds. The darkness drops again; but now I know That twenty centuries of stony sleep Were vexed to nightmare by a rocking cradle And what rough beast, its hour come round at last, Slouches towards Bethlehem to be born Using Midjourney
|
# ? Oct 2, 2022 17:05 |
|
portrait, the last emperor of the universe, by greg rutkowski and alphonse mucha
|
# ? Oct 2, 2022 20:11 |
|
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. Well two things, when Rembrandt did a quick drawing it was not like his paintings but very dynamic and deep form of sketching. It was stuff he did cheap and quick with his version of a ballpoint pen. Now what I think is interesting is these things native sketching styles, the way they look unfinished compared to more touched up ones. It's like they are blocking out forms of shadow on faces etc. Yet still there appears to be an invisible programmer, an artist interpreter to make the interpretation palatable for man. And on ai art I've seen that I like, I see people spending the same amount of time with the program describing it. And on the speed thing...frankly, I think I will be interested when we make a computer who can take his time to think. I will say, to a prismatic computer with an orange glowing light in the corner of my oom, to please give me a good visual hook for a production of the minotaur in the labyrinth, and three hours later it will come up with "I have my own life, I can't talk right now." That would be a great computer to work with, because he would have irrational individual taste. And it's not "toiling" it's fun to do! Why do you think data from star trek paints? It's leisure, it's fun. You don't need to automate this like complicated math. mobby_6kl posted:You don't have to actually do it. 3D modeling and drawing (in photoshop or otherwise) is pretty time consuming. Of course you can spend hours here too but generally you can write a prompt and get an image of pretty close to what you want in like a minute. You can refine it further in SD or finish in photoshop. This took a few minutes with a very basic prompt like halloween goth girl Why not merely invest in a cd of festive clip art?
|
# ? Oct 2, 2022 20:32 |
|
FunkyAl posted:And on the speed thing...frankly, I think I will be interested when we make a computer who can take his time to think. I will say, to a prismatic computer with an orange glowing light in the corner of my oom, to please give me a good visual hook for a production of the minotaur in the labyrinth, and three hours later it will come up with "I have my own life, I can't talk right now." That would be a great computer to work with, because he would have irrational individual taste. And it's not "toiling" it's fun to do! Why do you think data from star trek paints? It's leisure, it's fun. You don't need to automate this like complicated math. Was this trying to make a point, or a joke. Data isn't real btw
|
# ? Oct 2, 2022 20:40 |
|
FunkyAl posted:Why not merely invest in a cd of festive clip art? because it will cost money and won't be what I want. And also this was an example, I don't have friends to invite for halloween obviously
|
# ? Oct 2, 2022 20:43 |
|
I'm sorry buddy
|
# ? Oct 2, 2022 20:47 |
|
|
# ? May 30, 2024 09:47 |
|
|
# ? Oct 2, 2022 22:58 |