|
a pod system similar to Keurig that makes cups of steaming hot nacho cheese
|
#
?
Nov 30, 2022 21:13
|
|
|
|
| # ? May 30, 2024 16:48 |
|
|
AARD VARKMAN posted:a pod system similar to Keurig that makes cups of steaming hot nacho cheese The purpose is horrifying but that design is sleek
|
|
#
?
Nov 30, 2022 21:23
|
|
|
what happens if you eat all the watermelons on earth
|
|
#
?
Nov 30, 2022 21:32
|
|
|
|
|
#
?
Nov 30, 2022 21:38
|
|
|
edit: wrong droids
Cabbages and VHS fucked around with this message at 19:48 on Dec 1, 2022 |
|
#
?
Nov 30, 2022 21:49
|
|
|
TheWorldsaStage posted:The purpose is horrifying but that design is sleek I mean, outside of the pods this is nothing new. The Soviets went nuts for these things in the late 60s and early 70s:             ektachrome 35mm photograph of a 1970s Soviet kitchen device which dispenses liquid cheese, cup full of hot liquid cheese, product shot, promotional photo, product design, sleek design, Soviet 1960s design
|
|
#
?
Nov 30, 2022 21:50
|
|
|
I unironically would enjoy hot cheese sauce on demand...
|
|
#
?
Nov 30, 2022 21:53
|
|
|
Kharmakazy posted:I unironically would enjoy hot cheese sauce on demand... do not add foam and a straw but yes it would be nice to have on demand nachos.
|
|
#
?
Nov 30, 2022 21:58
|
|
|
i have descovered a fantastic new word of power, Vantablack!                   additional things Pineapple Hedgehog   A wedding cake made of meat pies   
|
|
#
?
Nov 30, 2022 22:22
|
|
|
|
|
#
?
Nov 30, 2022 22:29
|
|
|
This isn't directly AI-related, but I'm working on a low-animation-requirement game that's not too worried about internal graphical consistency, except that I'd really like world-space interactables to be blocky 3D. A long time ago there were some tools that allowed one to create simple 3D models from 2D images -- even a single 2D image if you didn't mind symmetry. I'd really like to use AI art to generate the base image(s) and use those to get a 3D model, but my 3D modeling skills are horrendous. Even with nothing but rectangular boxes. 1) Does anyone know of any (preferably free, open-source) software to convert 2D to 3D? 2) Does anyone have any tricks for getting multiple angles/shots of the same inanimate object?
|
|
#
?
Nov 30, 2022 23:12
|
|
|
feedmyleg posted:I mean, outside of the pods this is nothing new. The Soviets went nuts for these things in the late 60s and early 70s:  Here is a list of different film stocks to try. With people especially the Portra ones make a huge difference. More art than science! Fujifilm Velvia 100 (soft) Kodak Portra 100, 400 or 800 (iso's) Kodak Ektachrome 100 Lomography 400 35mm Film Cinestill 50D, 400D (filmic quality) Kodak Tri-X 400 35mm Film (black and white) Ilford HP5 Plus 400 35mm Film (black and white) Usually I generate cars:     
|
|
#
?
Nov 30, 2022 23:18
|
|
|
Was this one of those Midjourney word of the day prompt exercises recently? I went through a gallery of trending pictures the other day and about a third of them had "vantablack" somewhere in the prompt.
|
|
#
?
Nov 30, 2022 23:19
|
|
|
Hammer Bro. posted:This isn't directly AI-related, but I'm working on a low-animation-requirement game that's not too worried about internal graphical consistency, except that I'd really like world-space interactables to be blocky 3D. nvidia is working on some ai tools that can generate a 3d model from a 2d image but afaik it's still very much in research stages, and the AI model will probably be too big to run locally on a consumer GPU
|
|
#
?
Nov 30, 2022 23:25
|
|
|
what's the latest info on which sampling method/steps to use? Is there any good reason to swap off Euler A? I know the "a" ones scale very differently with step count than the non, but I see a lot of options I don't remember and wonder if there's a clear reason to swap
|
|
#
?
Nov 30, 2022 23:36
|
|
|
use one of the dpm++ samplers, they are the newest state-of-the-art ones and have the best combination of "gets a good image in few steps" with "runs pretty fast per step" euler_a is still great too though and i think is better than dpm++ 2S a if you're doing step counts around 12-20 the advantage of non _a samplers is that they are more stable so you can rerun a generation while tweaking the step count or batch size without basically nuking/re-randomizing the original results. if you don't care about redoing a generation with the same seed to tweak settings then just use _a forever because in exchange for that instability they get better quality images in fewer steps e: actually on my machine dpm++_2S_a runs at half the speed of dpm++_2M for whatever reason, so I guess just stuck to 2M and/or euler_a RPATDO_LAMD fucked around with this message at 00:38 on Dec 1, 2022 |
|
#
?
Nov 30, 2022 23:39
|
|
|
AARD VARKMAN posted:what's the latest info on which sampling method/steps to use? Is there any good reason to swap off Euler A? It's all so wild and chaotic that I can't actually tell if one sampler works better than others or not. Since most of this heavily relies on random noise (by design) it just depends on what you're after. I have a vague fog that I have in my head and I see what the computer fills it with. I suspect if you have a complete shape in your mind already then the samplers will come more into play. I treat this more like a slot machine with maybea cool picture on the end after I pull the arm.
|
|
#
?
Nov 30, 2022 23:40
|
|
|
KakerMix posted:
I made some vehicles recently too but yours are much nicer. Mine are literally garbage   
|
|
#
?
Nov 30, 2022 23:44
|
|
|
Crapple! posted:I made some vehicles recently too but yours are much nicer. Mine are literally garbage cool
|
|
#
?
Nov 30, 2022 23:46
|
|
|
Crapple! posted:I made some vehicles recently too but yours are much nicer. Mine are literally garbage extreme 'cool thing grandma has on her shelf that you want to play with as a kid but are forbid from doing so' vibes
|
|
#
?
Nov 30, 2022 23:50
|
|
|
Crapple! posted:Was this one of those Midjourney word of the day prompt exercises recently? I went through a gallery of trending pictures the other day and about a third of them had "vantablack" somewhere in the prompt. Lol yeah I'm a trend thief anything that looks cool in the trending pictures I steal the prompt
|
|
#
?
Dec 1, 2022 00:42
|
|
|
RPATDO_LAMD posted:nvidia is working on some ai tools that can generate a 3d model from a 2d image but afaik it's still very much in research stages, and the AI model will probably be too big to run locally on a consumer GPU I'm fine with a little bit of manual work in the model -> image stage if necessary. I just lack the skills to create the model myself. The existing technology (Adobe Canoma?) was impressive back in TYooLY2K. I feel like things have had to advanced since then, though maybe the concept never caught on.
|
|
#
?
Dec 1, 2022 00:45
|
|
|
Rutibex posted:Lol yeah I'm a trend thief anything that looks cool in the trending pictures I steal the prompt I just tried it for making night time environments and it seems to work quite well. Much better than my previous attempts which were never dark enough.
|
|
#
?
Dec 1, 2022 01:15
|
|
|
Hammer Bro. posted:I'm fine with a little bit of manual work in the model -> image stage if necessary. I just lack the skills to create the model myself. now I'm confused as to whether you want model to image or image to model
|
|
#
?
Dec 1, 2022 01:20
|
|
|
KakerMix posted:It's all so wild and chaotic that I can't actually tell if one sampler works better than others or not. Since most of this heavily relies on random noise (by design) it just depends on what you're after. I have a vague fog that I have in my head and I see what the computer fills it with. I suspect if you have a complete shape in your mind already then the samplers will come more into play. It's all about getting the right balance between speed and quality. Some of the newer samplers take less steps to stabilize into something usable but take a bit longer per step. Here are two huge grids I generated comparing samplers, the number being the number of seconds they took to generate:   Depending on the detail you need, sometimes you can get an usable image in very few steps.
|
|
#
?
Dec 1, 2022 01:43
|
|
|
in todays episode of Half in the Bag they used an AI voice changer to do a Rick and Morty bit  https://www.youtube.com/watch?v=_UbrN-wb0R0&t=517s this is what I think they were using: https://www.youtube.com/watch?v=gooIrcZEVSE
|
|
#
?
Dec 1, 2022 01:47
|
|
|
Rutibex posted:in todays episode of Half in the Bag they used an AI voice changer to do a Rick and Morty bit I think you're mistaken and that's actually Justin Roiland performing for real. They've worked together before on whatever VR game Roiland had. It's also completely on brand for those guys to have more-famous-than-them show up and they won't make a big deal about it. EDIT Yeah it's right in the credits. KakerMix fucked around with this message at 02:47 on Dec 1, 2022 |
|
#
?
Dec 1, 2022 02:44
|
|
|
I'm using the InvokeAI WebUI and they posted an announcement video for their upcoming 2.2 release. I'm not following other WebUIs that closely and the InvokeAI guys are in general a bit slower with their release/feature cycles, so maybe this is old news and boring, but this giant canvas thing looks neat. https://www.youtube.com/watch?v=hIYBfDtKaus
|
|
#
?
Dec 1, 2022 02:54
|
|
|
KakerMix posted:I think you're mistaken and that's actually Justin Roiland performing for real. They've worked together before on whatever VR game Roiland had. It's also completely on brand for those guys to have more-famous-than-them show up and they won't make a big deal about it.  holy poo poo
|
|
#
?
Dec 1, 2022 03:03
|
|
|
mcbexx posted:I'm using the InvokeAI WebUI and they posted an announcement video for their upcoming 2.2 release. The whole unified canvas thing seems like a pretty interesting take! Looks like it's focusing on more of a higher-effort multi-step workflow instead of just spitting out raw images. I've seen similar stuff in a few online tools (there was one collaborative one where anyone could come and add to the canvas with a prompt, I forget the link) but it isn't that common.
|
|
#
?
Dec 1, 2022 03:53
|
|
|
Sorry if it's been posted but https://twitter.com/rob_sheridan/status/1597692711314018306 Jeez. It's hard to believe these transformer language models came up within the last 5 years. I can only imagine what the next 10-15 will bring. Also I would be sweating bullets if I worked in commercial art
|
|
#
?
Dec 1, 2022 05:27
|
|
|
There is a paper about poisson flow generative models that work on a similar principle but work 10-20x faster than diffusion models, it will be interesting to see if those work out. But in 10-15 years there will probably be a lot of stuff that's way more gamechanging. The USA may or may not ban a lot of it due to copyright nonsense but I'm sure China will keep on kicking regardless. It's not like web scraping is impossible.
|
|
#
?
Dec 1, 2022 05:30
|
|
|
RPATDO_LAMD posted:There is a paper about poisson flow generative models that work on a similar principle but work 10-20x faster than diffusion models, it will be interesting to see if those work out. Very interesting thanks for the link. Sounds like there could be some huge legal battles ahead for AI.. and it's hard to know on which side of the legal coin AI will land. Will probably depend a lot on which administration is in power at the time or whether the Supreme Court takes on any big AI cases. I have to imagine that big businesses are salivating at the thought of replacing workers with transformer models.
|
|
#
?
Dec 1, 2022 05:36
|
|
|
Orc Priest posted:Very interesting thanks for the link. art generators are just an interesting side effect of combining AI vision models is with text models. the US government is absolutely not going to bother these AI nerds, and is instead giving them billions of dollars for their strategically important research
|
|
#
?
Dec 1, 2022 05:44
|
|
|
RPATDO_LAMD posted:now I'm confused as to whether you want model to image or image to model Sorry, I meant AI -> Image then Image -> Model. Though I suspect the latter would be greatly assisted if the former could output multiple close-enough images of the source content. In this first case, a treasure chest.
|
|
#
?
Dec 1, 2022 05:48
|
|
|
a vending machine that only serves cold beans before I upscaled it that was a single black bean 
|
|
#
?
Dec 1, 2022 06:43
|
|
|
AARD VARKMAN posted:before I upscaled it that was a single black bean
|
|
#
?
Dec 1, 2022 07:40
|
|
|
playing with some image prompts today. source image: "Soviet aqua troopers"               
|
|
#
?
Dec 1, 2022 12:21
|
|
|
Rutibex posted:playing with some image prompts today. source image: The earliest iterations of SEA PATROL.
|
|
#
?
Dec 1, 2022 15:57
|
|
|
|
| # ? May 30, 2024 16:48 |
|
|
KwegiboHB posted:Post that one.  the single black bean is also huge and weird but definitely not a person e: did you think I meant the whole image was just one black bean lol AARD VARKMAN fucked around with this message at 16:04 on Dec 1, 2022 |
|
#
?
Dec 1, 2022 15:59
|
|