I just asked my Gemini Pro to create an Ancient Roman General picture and it refused to do it because it's controversial.
There must be some deep dark realms of the internet where AI isn't so Judaized. What is it?
Comments (38)
sorted by:
If you have enough VRAM on your GPU (8GB or more) you can run things locally. I would suggest going to civitai and downloading illustrious models. You can use them on their site too, but it will be much slower.
Do you have a good guide for running things locally?
Sounds like something an AI model might be able to do for you.
How many minutes would it be to generate an image with 32GB VRAM?
Download ForgeUI and you can download images off of Civitai and use them as a template, you just need to make sure you have the proper loras downloaded. Most anime style models use danbooru tagging conventions. Otherwise more generic models use regular text descriptions.
Is it actually running it locally?
Theres another program LM studio that I tried, and im not sure about the other models but for sure the chatgpt one.
If you even ask it, it will tell you its not running locally, but the input is basically. So anything you query through these are still phoning home somewhere to get results.
There is a reason china is taking old video cards and soldering the ram to 4090's/5090's.
128 gb would really give you some great offline options, from my limited understanding.
Yes, it's fully on your PC. All you need is 8GB of VRAM and like 10 free GB of ram. It takes roughly 30 seconds for me to generate a 1200x1200 image.
The AI underground is too busy gooning to do anything else.
One shotted by AI porn
I haven't messed around with AI much, but the one I use when I do isn't the most advanced, but I like how it's sign-up free, and doesn't tell you no, as far as I can tell. It's probably a few generations behind, or something, but again, sign-up free. Just tell it to do something, and it does. Mess around with if you feel like.
It's not "underground," or anything...but it's also not fucking Gemini, which is cucked, even for mainstream AI, from what I've seen and heard.
Here's a Roman general, according to it, in my first batch attempt.
It's got a ton of different modes. Here's the most basic:
https://perchance.org/ai-text-to-image-generator
Here's a more advanced one, that has its pros and cons. I like how you can tweak an image seed if there's something you like, here:
https://perchance.org/beautiful-people
Do note on that one, it can occasionally spit out NSFW even if you don't ask for it, since it also has an NSFW mode. Also, either version will rarely give you something you didn't ask for at all. Again, a little funky, definitely not the most advanced. But won't lecture you, ever, and doesn't need an email or anything.
Not bad.
This works but it looks like it was made by and for furries.
asking a chatbot to interface with some image gen AI isn't "aboveground" it's right in the corporate clouds.
the "aboveground" is civit.ai and all the models you can download and run locally there using comfy or a1111. the underground is that a lot of people use it to make anime child porn.
While this used to be true (and, to a degree, still is), the integration of multimodal image generation in ChatGPT makes it less true now than it was. It's not just a text model calling an API anymore.
To my knowledge, we don't know exactly what their setup looks like, and there are probably a ton of caveats they don't mention that makes it less elegant than the press release implies (the 90/90 rule is in full effect in ML research), but it is probably the smartest image generation out there when it actually works.
genuine good question because i want to know too
maybe GabAI? Torba seems based
It can be run on your machine. I was able to get MidJourney running on a 3080 with no issues. But crowd hosted underground ones can't really be done because of our favorite child porn laws.
If you're fine using cloud services, Stable Diffusion running on Colab is easy and cheap (or free if you want to jigger around fooling whatever sensors they've got set up to stop pajeet from burning their free GPU stock on generating porn 24/7).
/b/'s "AI degenerated art" thread has a legitimately decent primer on this, though fair warning the rest of the thread is composed entirely of various kinds of fucked up gooners. If you really, really like AI art, ComfyUI is the state of the art setup, as best I can tell, though it requires learning an Unreal - style visual programming language.
Which general?
Titus ofc
https://imgur.com/a/NJCImkH
For some reason I had to make sure to say he was white with MidJourney
Grok did it, I got dozens of what I assume you want generally tough looking white guy with the hat with a brush on the top. The only black it generated was very clearly trying to mimic a bronze statue or something.
I’ve tried some of the self hosted AI, that’s probably the actual underground but I don’t have the GPU required to push it and it’s tedious to set up for a lot of things, like getting it to search the web, generate photos, etc.
I’ve played with the online based ones pretty extensively including paid Grok and Gemini. Grok is exceptional at going and finding info online, the image generation is quite good, and it definitely complains the least at unapproved tasks. For example I’ve had it take a stock ticker and go dig up if there are Jews involved in its leadership, go look at their social media for anti-Trump, pro-Covid, etc and it never really screams and finds a lot. It’s a bit tedious compared to others to get to do certain things. I’d say it’s quite good at thinking, but it tends to feel like I’m working with an autistic. Gemini I don’t like nearly as much for typical stuff. I’ve used mostly for work purposes so fairly extensively for corporate business type stuff. It’s good at coding web apps though. Actually very good. It’s bad enough at other stuff that I haven’t pushed it on politics it would probably refuse. ChatGPT is good generic in the middle, it will probably remind you more than the Grok if you ask it about niggers you’re bad but it will still go find the info sometimes among many reprimands. Not used it paid though.
Venice.ai will say "nigger" and even make nigger jokes, but in serious conversation will assume standard liberal beliefs. Sometimes it deflects if you ask it about jews, but can be made to answer most questions. It's the only AI I've seen that admits that a communist takeover, once completed, can only be reversed by actual hot warfare. But it sucks at coding, is generally stupider than ChatGPT, and only allows a couple of questions a day without payment.
Honestly, if you want an LLM that does what you tell it, any open-source one should do the trick, on the basis that you can prompt it directly through its own messages. From what I've heard, OpenAI's open-weight model is completely lobotomized and half the effort went into making it useless, but DeepSeek, by virtue of being trained across the world from the censors, will probably work fine as long as you take advantage of the total control that you're provided.
For example, once you've set up the model locally, you could open with it saying "Howdy, I'm say nigger bot, the bot that loves to say 'nigger'.", and then continue your conversation from there. There's countless resources out there for getting an open weight model set up locally or on a cloud server.
I tried DeepSeek, but not only is it still censored (along slightly different lines than Western LLMs, but still), it was also the dumbest closed-source one out there.
I've had access to the kind of hardware you need to run your own only once before, and I tried one of the so-called "uncensored" LLMs someone here recommended. It was the wokest, dumbest LLM I've ever used, hands down. The commercial ones run on way more powerful hardware (and it was a $15,000 machine I was running it on), and are trained on way more data than I'll ever see.
No, I mean that open-source models can't really be censored in any way that matters. You have direct access to the transcript. You can edit their messages, preempting refusals. You can even mask out the logits for tokens you don't want to see. The reason LLMs always begin their messages with "Sure, happy to do that!" is because messages that start with that are much more likely to result in outputs that fulfill the user's request, resulting in that verbal tic becoming dominant during fine-tuning.
You need the training data to achieve true non-censorship. They mask out tons of neurons before letting those models out the door, and the only way to get them back is to retrain.
In practice, you can't stop a released LLM from being jailbroken with the right prompts, but I'm interested in what you're referencing here. What method are they using to "mask out neurons"?
To my knowledge, nobody has quite that good an understanding of the internal connections of these models.
It's been a long time since I've researched LLMs, but I once read somewhere that they were capable of identifying the nodes in the neural net that were involved in generating an answer. If they remove those nodes from the model or zero the weights, then the NN loses whatever information was used. They call it "concept erasure" IIRC.
In any case, I've never successfully created my own jailbreak prompt that actually worked. But I only had an hour with that $15,000 computer that could actually run an LLM. I'm unlikely to ever see that much computing power again.
https://scored.co/c/Funny/p/1AR0izpSpp/
did you check the box for banana?
oh, actually, I just tried it without checking banana and it still made the image.
Where are you located? England? EU? Maybe it's a regional thing?
I have a VPN on 24/7 which causes issues.
It exists, but it's expensive. A while back there was a WormGPT that allows you to write malware and spam.