Does anyone know how to recreate DALL E?

I kinda want a DALL E but the AI can search for explicit photos for reference. I'm a coomer, I'll admit this is for dumb coomer reasons.

In Dall E mini, I can more or less see the limits of what the AI was chosen to create. Pretty much as long as it can open up on google images safe search, it can load.

This is just an idea by a coomer, just a dumb little idea. Since Dall E 1, isn't as smart, we won't get as detailed stuff but I think that's fine. I'm just kind of curious to see exactly what it can make.

If someone does see this who can bring this idea to life, idk, put ads on it and make bank on it idk. I just kinda want to see it exist.

Attached: soyjack_dolly_meme_template_by_dabigboss888_df0s802-414w.png (414x280, 107.06K)

Other urls found in this thread:

youtube.com/watch?v=eWpzLIa6v9E
twitter.com/SFWRedditImages

ur not smart enough to code it

Yeah definitely not, the most I'm versed in is rpgmaker but even a kid can use that.

It would probably be easier to just drop the porn addiction

Run dall-e mini locally, train it on rule34 or whatever

How do I do that? I downloaded dall-e on a zip file, but how do I train it?

Dall-e 2 and Imagen both have papers that show how to reproduce them. Your best bet is to start with "Deep Unsupervised Learning using Nonequilibrium Thermodynamics" then move for the foundations on diffusion models, then check out "Improved Denoising Diffusion Probabilistic Models" for a useful implementation on image creation.
As for training data, Danbooru has a well annotated dataset for machine learning\

This is not for brainlets though, it's a lot of work.

This, you can train dall-e mini yourself

typical braindead coomer, nevermind I guess

>then move for the foundations on diffusion models
*then move on to the foundations of diffusion models

While I'm not super serious on this whole idea, just thought of it on a whim. I'll give myself some time to learn it.

If I'm going to give the dall e clone some training data though, maybe not rule34. Perhaps just google search without safe search as I think that's a little more wide.

just a friendly warning: if you plan to use something that uses online databases for cp, you are an idiot

For real, don't open pandora's box. It's the last thing the ML field needs right now.

defintiely not, i'm not interested in that type of stuff
you have my word on that. (source: me) my tastes surround milfs and stuff

Have you tried reading the docs you fucking numbskull

hahahahah

I have, but I'm a tard so I can't make sense of it yet. Still reading the Deep Unsupervised Learning using Nonequilibrium Thermodynamics, but it'd prolly take a month before I actually make sense of it.

It would be easier to learn how to draw. I believe a basic anatomy course with some practice copying hentai artworks will get you coom worthy material in a month training maybe an hour a day

I actually have learned how to draw pretty much all that I want, this whole idea was just a shower thought.

Not too serious on it.

Don't sweat perfectly understanding everything. If you can get the basis of the process, you can probably implement something.

Actually, I'll stay a numbskull and just think about this. If someone's interested in this shower thought of mine they can go ahead and make use of it.

This'll probably pop back up in the back of my head one day and maybe I'd get the motivation to do it.

In a few months I'm entering college, and the course is game development so I can learn about programming and art stuff.

I'll give myself a week, if I can't make sense of it. I'll just leave it to the minds of whoever sees this.

GL user and stop self-ironizing so much

Good luck. Protip: Math courses suck, but you need math if you want to do anything really cool/cutting edge.

I'll start with this video.
youtube.com/watch?v=eWpzLIa6v9E

Just train it with cooner pictures

Thanks, and my bad on the ironizing part

There is no running dallee onyour own pc
Just running dallee on the cloud
You better shell out some money bitch

Who dis?

You might be able to run inference, but training/fine-tuning a model of that size is out of reach for a consumer PC. If you want to generate anime porn you'll need to make your own.

damn

Nobody popular in particular, just an average coomer who can draw lewd stuff on their own and is studying game dev