- Joined
- Apr 22, 2015
Yeah, thank god for this thread for research purposes because this is one of the few non-faggy avenues to try to learn this shit.
Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump
- Thread starter Near
- Start date
-
🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
- Joined
- Dec 18, 2022
My Advice is to check out the Linux thread. Get Mint instead of Ubuntu if it's your first Linux and are migrating from windows. Then install the Cuda packages for the compatible Ubuntu Distro.
- Joined
- Aug 6, 2021
I do wonder how animations will end up.
It does feel like we are at the early stages of personalised media.
I have tested luma with a basic animation but I've seen some crazy stuff posted around
It does feel like we are at the early stages of personalised media.
I have tested luma with a basic animation but I've seen some crazy stuff posted around
- Joined
- Aug 17, 2018
Back again from Bing's image generator.
It accepted actress Doris Day! Eating a burger with a dog.
If you could describe the 1950s in one word...
It accepted actress Doris Day! Eating a burger with a dog.
If you could describe the 1950s in one word...
- Joined
- Oct 20, 2019
I saw you have AMD hardware, I do too. If you had Nvidia I'd suggest just doing everything on Windows which you already have, but for AMD you need Linux for it to be performant. The reason is because doing it on AMD requires a specific AMD set of libraries called ROCm and the Windows version is way behind the Linux one. If and when it catches up, you'd be able to do it on Windows just as easily but there's no imminent chance of that happening.
That said, the software will work on Windows. There's nothing stopping you from just trying out Automatic1111 or ComfyUI on Windows. You'll just be waiting for ten minutes for your image rather than 70 seconds.
When it comes to your choice of how to do this, Automatic1111 is generally said to be simpler than ComfyUI and I guess perhaps it is.
You'll need to install the ROCm libraries on your Linux box. There are a few ways you can do this but I personally would recommend just following this: https://rocm.docs.amd.com/projects/install-on-linux/en/develop/how-to/native-install/ubuntu.html
I can possibly give you some help if you get stuck. For choice of distro, I prefer Mint but you might want to pick Ubuntu just to minimise differences when you're following How-Tos. It's not like you can't change distro later.
In the words of Dorothy Parker: "I knew Doris Day before she was a virgin."
indigoisviolet
kiwifarms.net
- Joined
- Jan 18, 2022
Aaaand... it's over as soon as it began...
Just in case you weren't unimpressed enough by Stable Diffusion 3, turns out they put a whole lotta legal mumbo jumbo into the EULA that has forced civitai et al to suspend sharing of anything touching it for the time being.
It could all resolve tomorrow, or it could go on forever. Either way, you'll prise my beloved SD 1.5 out of my cold, dead hardrive.
Just in case you weren't unimpressed enough by Stable Diffusion 3, turns out they put a whole lotta legal mumbo jumbo into the EULA that has forced civitai et al to suspend sharing of anything touching it for the time being.
It could all resolve tomorrow, or it could go on forever. Either way, you'll prise my beloved SD 1.5 out of my cold, dead hardrive.
- Joined
- Dec 9, 2017
I will certainly miss all of the ‘girl laying in grass’ images.
To that note, did anyone have any good experiences with SD3? The only thing that impressed me was its ability to sanely generate text where I prompted it. Photorealism wasn’t any better or worse than what can be achieved with checkpoints like Juggernaut, etc. Cartoon/anime characters and scenes were also pretty good but like others have mentioned, no reason to use SD3 over whatever you’re currently using.
Have any of you trained/used some DoRAs? Thoughts?
To that note, did anyone have any good experiences with SD3? The only thing that impressed me was its ability to sanely generate text where I prompted it. Photorealism wasn’t any better or worse than what can be achieved with checkpoints like Juggernaut, etc. Cartoon/anime characters and scenes were also pretty good but like others have mentioned, no reason to use SD3 over whatever you’re currently using.
Have any of you trained/used some DoRAs? Thoughts?
Last edited:
Potatoherder
kiwifarms.net
- Joined
- Jun 28, 2020
And they completely fucked up with the naming. No one will make the distinction between SD3 2B/4B/8B. It will forever be known as a shit model with a bad license.
- Joined
- Dec 18, 2022
- Joined
- Jun 3, 2024
Took 6 minutes to generate, probably because it worked with a screenshot from a ye old PS eye vdeo.
- Joined
- Dec 18, 2022
Welp SD3 is unusable for the coomers. It's DOA. Pony is apparently staying with SDXL. All of the really good coomer finetunes have commercial support. They essentially sell generation as well. With SD3's licensing no serious finetunes to rival SDXL finetunes are going to come out.
- Joined
- Oct 20, 2019
You... fiend. How could you? Look what you did to my girl. *sob*
Sure. Kind of. The API version has been confirmed to be running the 8B version and frankly it made a strong first impression. Prompt adherence and quality of output. And on Reddit you can find a number of posts where there are comparisons between the local version and the 8B API version. It's far superior in both understanding prompts and in quality of output. I could do you a few comparisons if you have any requests.
If you mean the local 2B version which of course I'm fully aware that you probably did, then somewhat. It understands prompts well, it ran quickly on my hardware and for a base model with no fine-tuning, I found its output impressive. It's not bad, it's damaged. They did that to it. It should be good and you can see that it would be from when it works.
- Joined
- Dec 9, 2017
- Joined
- Dec 18, 2022
That ptx0 guy is the only one talking sense. Everybody else in the comments thinks that Pony is the only thing keeping Stablity AI afloat. I think Stable Diffusion is their only good product though. By good, I mean midrange in the industry. Midjourney is better by most accounts.
Stable Beluga is a joke LLM that is just a bad Illama 2 finetune. With Illama 3 out a month or two ago it's over.
Kaiba Cuqk63
kiwifarms.net
- Joined
- Jan 6, 2022
This is why I install anaconda alongside system python, so you can use multiple versions. You should try to avoid updating python until your current version is running behind, imo because then you're stuck figuring out what doesn't work anymore. All these different tools and a bunch of them are very strict about what python, torch, and cuda version you use. Also if you install all your dependencies in one go in a conda environment there are fewer conflicts.
- Joined
- Dec 3, 2020
Was bored on civitai and decided to remove the NSFW filter.
First thing I see is a dragon fucking a car...
Then again I wasn't particularly looking for it so who knows...
First thing I see is a dragon fucking a car...
Stability is dead now, sucks tho since all other options at the same level can't be run locally. For all the hoopla about SD being used to make CSAM I've yet to see any. Seen plenty of weird porn specially furshit (barf) but no actual photorealistic cp anywhere.
Then again I wasn't particularly looking for it so who knows...
Closed source SaaS shit, hard pass, shame that the same guy from leap motion would go with this model given how open that thing was.
- Joined
- Aug 28, 2019
Runway’s latest AI video generator brings giant cotton candy monsters to life (archive)
Nothing publicly available, just a demo.
Nothing publicly available, just a demo.
std::string
kiwifarms.net
- Joined
- May 17, 2018
IWF (PDF link) did a report about it
Under UK law it's just as illegal as the real thing (assuming it's photorealistic) so that might drive down its appeal a bit. They're using checkpoints and LoRAs just like the furries and bronies do.
- Joined
- Dec 18, 2022
Wow. Turns out if the cops find you with an indistinguishable copy of something illegal, it can be used to convict you.
People who generate facsimiles of CSAM, if it were legal to do so, are doing a great job of heading towards a possession charge.
- Joined
- Jun 23, 2016
Ran my avatar through Luma AI.
Last edited: