The Dall-E AI Has Gone Too Far... 

Use code "SOG" at www.gfuel.com and save some money. Check out the newest podcast episode: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-Cwl2NBI4xzE.html

I can’t stop laughing at the discord moderator one, this AI is so based and I love it so much for that

If anyone wanted to make a horror game out of horrifying generated photos, Dall-E would be the best option.

The fact that 2 months after the Dall-E 2 AI paper had been published, google published a paper on the their Imagen AI which is even better than the Dall-E 2 AI really speaks volumes of how fast technology is progressing.

The mini version is trained on unfiltered content that's why it's generating edgy stuff, the main Dall-E (not the mini) is trained on a bigger and filtered data set

There's a lot of oddities with this AI, generating "November 5th 2024" leads to images of people lying dead on the ground and "The End" is just a shadow figure in a dark empty void.

"in the future, entertainment will be radomly generated" I feel the writers of vegitales knew alot more then they let on

As a professional environment artist, Nvidia Canvas and these AI NLP image generators have helped me a lot when I am struggling to come up with a concept for a client. I can type in what the client wants, or paint in a basic scene in Nvidia Canvas and boom! Art block is over and I have a low res concept to model my environment from. This is far too sloppy and uninspired to use as a an actual piece, but it helps me to get the ball rolling with a general idea when I have multiple clients with tight deadlines. I can even make quick changes and send over precomps to clients, spending maybe an hour in the idea phase rather than days. A lot of people do not understand that most clients request an 8K or 16K final render and an AI would take years to compute a 16K image, even with the most powerful hardware. Even if it was faster; style, composition, multiple angles, and separate elements are nearly impossible with AI. You can say "In a realistic style" or "In the Style of Picasso" but the AI can only search the internet for images with those tags and attempt to apply them. It can not replicate proper brushwork, precise lighting, proper SSS, Realistic AO, global illumination, etc. It can try, but you will spend more time re-rendering a high res AI image over and over than the time it would take just doing it yourself. AI is great for visualization and fun, but terrible for anything else.

An interesting take I heard on this from Niko at Corridor is that, by their nature, AI cannot create new art, they can only remix existing data. They might be able to make a convincing image, but it's up to the artist to either tell them what the image is supposed to be, or to manually make it the way the artist desires.

Imagine an entire film being created with this tech. All AI-generated scenes, voices, sounds, script and music. Christopher Nolan would have a heart attack.

7:55 sorry muta but that is 10/10 accuracy

it's a pretty good AI, it can be used to generate characters with a description or just make Jimmy Neutron crossover with literally anything imaginable

Even though they tend to look like a Rorschach inkblot test, it’s still really impressive how it can create anything from a sentence

not only is it a really good ai, but its an ai with good humor

DALLE · 2 is much more powerful at generating images and faces overall, although it struggles with text. But its access is restricted to a small set of users. DALLE mini is not an OpenAI product, but everyone can use it. I though the distinction is important and I hope in the future you can get access to DALLE · 2 and make another video. Your inputs on that would be nice to see ;)

7:20 that play on words got me dying, "it doesnt stretch beyond any limits", in reference to sonic goatsie generation. cant tell if it was done on purpose or not

8:05 "the Ai has a weird perception of what it considers to be a discord moderator" You're right it's missing the furry graphic tee and it should be chasing a child through the woods. But hey it's getting close, props for that.

EDIT: The prompt was MOST certainly "Fat human trail cam", got the same image nearly every time as opposed to zero times trying "discord moderator trail cam" over 15x ran "discord moderator trail cam" 15x, never got anything close, even added fat to it. I wouldn't doubt that image is photoshopped unfortunately, just in the sense the actual prompt was "fat human" or something. That, or the a.i. is learning lmao

"It sort of doesn't stretch beyond any limit" Excellent choice of words lol

"Muta if he could sleep normally" "Charlie if he was 5'10"

If you haven't already Muta you should go over The Act Man situation again, because there are a lot of terrifying implications for ALL youtubers. Please cover it again as I think it is one of the most important topics going on right now.

4:58 Mutahar: "I'm not a big anime viewer" Also Mutahar: *Watches Family Guy* (A classic anime)

8:34 holy shit it generated SCP-096

There will come a point when AI generated images are imperceptible from an actual photo, at which point I think certain safety measures should be implemented. Such as a mandatory watermark or perhaps something written into the code of the actual file or a specialized mode of saving the file. Any way for an expert to be able to look at the code or the image and designate it as a generated image.

if there was a horror game based off of trail cam footage of a discord moderator, everybody would be too scared to play it.

Dall-E gotta be the most fascinating and scary AI image-generating program, 'it's honestly crazy how technology has advanced to the point of creating amazing and terrifying art.

8:10 We did it boys! This is the future.

The Ricky trailcam got me dead 😆

I swear Disney needs to stop with these Reboots. First Lion King and now Wall-E?

It won’t be long til DALL-E gets to be used for producing rule 34 arts…😏 But joke aside, it’s pretty impressive how the AI technology has evolved to the point of creating impressive art. The question of whether it could have creativity of its own like human artists instead of replicating already existing works is up to debate.

I lost it when TrailCam Ricky Berwick came up

muta mixes up his words from time to time in the funniest ways. "in a nutcase" instead of nutshell😂

I don’t know to explain this, but whenever I see ai creating images, music, or anything in art in general, I feel so happy and proud of the ai.

Seeing AI art go from Wombo, which is unrecognizable as anything at all, to Dalle•Mini, which can actually be discerned as what you inputted the AI to make art of, is astonishing to me. I can’t wait to see how this advances in the future, with limits in place of course.

You can't even use Dall-E mini at this point with the amount of traffic there is

I never could get past the fact that AI always seemed to just interpolate between data sets, for example, if you're using some AI face generator thing and set it to have a brown beard - it's literally going to use some dude's real beard and basically just blend it into other elements. So all those elements, are from real photos of real people, just cleverly blended. So it wasn't that impressive, as it's just not really generating new faces, other than a fancy Identicon generator that randomly chooses from a set of noses, eyes, hats etc. But DALL-E doesn't seem to be too reliant on its input data set, at least if it is, it's really good at hiding it. Really cool stuff.

7:37 "It has not however, created the blue hedgehob holding is wide asshole agape...I mean, standing next to a goat..."

dall-e-mini has a very limited dataset. So in most cases it'll just recreate a mutant version of certain pictures it's learnt from, or a combination of 2 or 3 of them. And it absolutely mangles faces, seemingly deliberately.

I just want to know what Muta keeps in his GFuel mini fridge lol

I was waiting for you to make a video on this

This AI is honestly perfect for weirdcore/dreamcore. Not great for other things, but excellent horror AI.

Yes, Sonic Goatse has STRETCHED the limits of AI... 🤪

I tried getting Dalle e to generate Mutahar as Commander Shepard, but sadly I didn‘t get Mutashep.

"It does not stretch beyond any limit" when talking about goatse. Couldn't have said anything more perfect.

SomeOrdinaryGamers is an anime channel now according to DallE

Thanks for the reminder, I need to set my alarm for tomorrow! :D

Lol I just made a short about dalle-mini... that shit is terrifying!! I got a bunch of pictures months ago before 4chan crashed the damn thing. Great video bro!

The best and the most terrifying thing about Dall-E AI is that it created it`s own language in order to handle NLP, without the help of scientists, completely on it`s own!

I love how the internet gets a hold of this super powerful and advanced image-generating A.I and immediately starts using it to make memes and cause chaos.

The issue is I can't get any porn or nudity out of it, no matter how hard I try

the AI has unfortunately become too popular and now its impossible to use because of too much traffic

DALLE has helpes me so much i was late on a photography assignment on architecture and shit and this helped me out soooooooooooooo much ;)

If there's one thing i like to do with Dall-E, it's having it generate anime characters into Breaking Bad (or with Walter White)

Im actually generating a dall-e mini prompt right now and man this is actually incredible

Love messing with this stuff, as an artist it can really help get the ball rolling during some art block. Hmm I wanna draw mario and herobrine partying in Minecraft? Here's a bunch of possible ideas. Ofc it doesn't do everything which is actually a plus, it just gives you some vague ideas that you can go "oh I like that layout" or "that's a cool idea" and then work with it. Idk if I'm putting this into words well XD but it's neat

The fact that you said it didn't "Stretch beyond any limit" with the sonic goatse thing gave me the hardiest kek I've ever had

Obviously with more data, the AI can more accurately depict the Discord moderator as fatter

I'm really bad at brainstorming ideas, and this program has been an absolute blessing for generating stuff like environment concepts for me. It's basically feeding you a prompt that you can then go and expand upon.

Gretings, Mutahar! Congratulations to your face for making it into a thumbnail of one of those text-to-speech reddit Story Channels ("UE Stories", in this case) with a video titeled "Spoiled Kids Getting Owned"! This must be every youtubers dream come true.

Im surprised you didint mention the way more powerful dale 2, that one truly blows my mind, reddit has some generated pics which make me have an existencial crisis

Its as funny as it is scary. Art and music defines us as a species. Its our special thing, something emotional and illogical but its essence was pure "human". Now though our contraptions have caught up to us and were losing what has made us unique. it's scary, it honestly is.

As far I know Dall-e doesn't generate faces well on purpose, so it can't be used to create questionable pictures of real individuals.

I love this ai only recently discovered it, i tell you seeing spiderman rob a store on ctv and tripping over in the next image is perfect

you definitely shouldve mentioned Dall-E 2, the images it creates are insane. i especially like the "orange cat staring at drawer of socks on fire"...... gramdson.....

Servers are lagging so much. I can't even generate one picture.

Wow that trail footage option is terrifying. That's so neat though

Having been a subscriber of Two Minute Papers, I got the scientific explanation months ago. Now for the Muta explanation!

Thank you, I literally searched this by chance today, before you posted the video. No idea this AI generation was getting traction. Truly, we live in interesting times.

Imagine the if the AI said " laughing my codes off" or simply "lol"

As an artist, I am freaked out. We were supposed to be the ones AI would never reach, due to creativity being hard for AI. And look, we are among the first ones being fucked.

I think a good use case for this tech would be in concept art for movies or games. Imagine how much the creative process would be streamlined if a writer could just feed a prompt from their imagination to this AI and get something interresting.

one of the funniest things i been seeing using this involves courtroom sketches and im losing my shit when i see it

I feel like with how exponentially fast technology is advancing overall as well as the art it already has the skills to generate, somewhere in the near future we will probably have whole ass AI generated games and movies.

AI has gone too far, but I hope it keeps going.

Imagine NLP based inputs for photo editing. Once you’ve got an image that it’s generated. If you can’t get more granular with the detail, then instead ask it to add or move things and it edits the image you’re talking about. This image being something it once generated.

Remember this is a super watered down version we have access to

Those discord moderators are historically accurate. A more realistic AI generated image will never be made.

I'm surprised nobody has attempted to corrupt DALL-E's Image results similarly to what happened to Tay AI.

Heh. You said "stretch" in relation to goatse. Well done, sir. 👏

I spit out my coffee when you pulled up Discord Moderator Trail Cam lmaoooo 😂 Omg 😆

I can finally have pictures of characters like Master Chief or Solid Snake having a vacation For some reason, it does generate Rosalina and Meloetta well

8:43 no, i didn't turn on my lights i promise

7:50 that reminds me, remember the human "cent-iPod" on Southpark? hahaha

7:27 "Doesn't stretch beyond any limit." I see what you did there

You gotta love AI technology, it always keeps on advancing with each second.😺

It's like someone took the idea of the Mandela Catalogue and turned into an AI. Every picture makes me think that Alternates are slowly taking over the world and warping the landscape as they do.

Those trail cam footage images had me laughing the hardest I've laughed this year

Dropped that twewy reference! Love it

IDK, I think DALL-E was fed more than enough information to create the perfect discord moderator.......

3:48 I am freaked out, because whats the point for artists in the future to keep making art if a robot can do it 1000x better in a few seconds? why even put in the months of hard work if it's pointless with these image generators?

The Rickey burwick one had me crying, I haven’t laughed that hard in a while

It's really nice that I beat my first MGS game recently, with it being MGS 2 and Muta has referenced it multiple times lately Yay

Oh god, it's happening! Everybody stay CALM!!!

As AI improves at generating fake images, sounds, or music, the more I question “is this really necessary?” Don’t get me wrong; I can see AI’s usefulness in everyday life; space travel, data sorting, facial/voice recognition, easy 3d asset creation, even checking CCTV for signs of a break-in. But can someone please explain if this might have negative side effects, too? Will photographs stop being viable evidence in court? Will certain design jobs be lost?

I spend my weekend on a Dall-E and you drop this masterpiece on me, spooky

I mean the big meme laugh was during the video where you looked at some horror site with an anime website as it’s backdrop so maybe that’s where the connection is?

My issue with current A.I is the utter lack of backbone the creators display. How many times have they caved and completely scrubbed data because of public backlash. Whether I agree with what the A.I is learning or not, I feel that the only way we are going to advance is by allowing completely organic learning with the absolute minimum of developer input outside of bug fixes and upgrades. Let it learn organically instead of erasing things and pretending they don't exist.

Dall E Mini shows us how far we can take art

I would be interested to see if we took book descriptions of scenes or people and see what it pops out with

Somehow Muta always get to talk about Metal Gear and i love it