Spaces:

dalle-mini

/

dalle-mini

like 5.43k

Running

Better but scary

hahahaha what the hell

Photo of Japanese woman performing a Shintoism ritual, wearing shrine maiden clothing in an empty shrine, at midnight, illuminated only by candlelight. + The equation of this post:

noiiice

How did you get the idea to make this "equation" by the way? I'd never imagine Dall-E understands emojis. It's amazing

almost there but still a lot to do here.
i tweaked it slightly for better results. try this one out
Artstation style = high detail : ( subject =highly detailed💃🏻🧕🏼+(anatomically correct facial features + (highly detailed = 👩🏼‍🦰🧝🏼‍♀️)+((highly detailed and anatomically correct (realistic and highly detailed + anatomically correct shaped eyes=👁👁),highly detailed and anatomically correct👃🏼,highly detailed and anatomically correct👄)))

only one of these is haunting my dreams tonight

they are all cute

I think it's just more about the "highly detailed" "anatomically correct" "anatomy" etc descriptors. Also, the artstation ones naturally have more source to draw from since most of the subjects in there deal with very detailed and higher res faces. Just a thought.

yeah but it does react to specifications of either subject background or ground and stuff like that.

would you mind posting your results if you have tried it out

Yeah I'll try the subject, ground, etc. I'll stick to the more natural language descriptions though, like JophASL, as that's kinda the point of the AI imho. Being able to parse a natural and detailed sentence and create context from that everyday usage. Good to experiment though, so keep at it.

How did you get the idea to make this "equation" by the way? I'd never imagine Dall-E understands emojis. It's amazing

I saw some one on another app input emojis randomly and I noticed that the AI reacts to them so I started putting them in intuitively and combining it with words. Why the equations? I have no idea all I thought is that its a computer programme something in it must be behaving the same

I do like the "anatomically correct". "realistic lighting" is a good tag as well. I try to think of the contextual activities that would naturally have examples of more detailed faces and use those as descriptors like, headshot, headshot photography, realistic, realistic lighting, etc etc. Honestly, there are just so many variations it's great to have all these tests. I'm using well know celebrities as there is a ton of material to use as ground truths to guide and hone the results.

So you can see #2&#3 are worse compared to the lower ones. #4 has a few really good ones but also she has very distinct eyes so the source material is much better for her.

Yeah I'll try the subject, ground, etc. I'll stick to the more natural language descriptions though, like JophASL, as that's kinda the point of the AI imho. Being able to parse a natural and detailed sentence and create context from that everyday usage. Good to experiment though, so keep at it.

I'm gonna be honest im not well versed with the whole programming thing so I don't even know what JophASL means, and I do think its good when the comp can understand simple inputs, I just tend to wonder how accurate can it be. As you can see their are a lot of limitations on the dalle- mini and im just looking for ways to get past them till they release the better version

I do like the "anatomically correct". "realistic lighting" is a good tag as well. I try to think of the contextual activities that would naturally have examples of more detailed faces and use those as descriptors like, headshot, headshot photography, realistic, realistic lighting, etc etc. Honestly, there are just so many variations it's great to have all these tests. I'm using well know celebrities as there is a ton of material to use as ground truths to guide and hone the results.

wow I really love it. My thing is I'm trying to synthesis faces that don't really exist, from scratch so to say. And as you can see there are limitations like details with eyes and all of that that im trying to over come with the equations and combinations as you can see descriptions only go so far. I also did the celebrity thing at first so I totally get you

I'm trying to pin down where exactly and how the interpretation takes place so that I can tweak it how ever I wish to without relaying on the black box effect of this AI thing.
There is a distinct language taking place and I'm really curious to figure out how and why it listens.
this is the current equation that im still tweaking for detail, but so far I like the response its giving

greg rutkowski style with highly level of detail: subject =🐱👧 in red dress - subject detail: anatomically accurate and higly detailed face=( highly detailed and real blue eyes👁👁 +highly detailed red lips👄 ), ground=🌷🌺🌿blush and vibrant

"My thing is I'm trying to synthesis faces that don't really exist, from scratch so to say. And as you can see there are limitations like details with eyes and all of that that im trying to over come with the equations and combinations as you can see descriptions only go so far."

Ah ok gotcha. Yeah I guess you're doing the same thing, but instead of describing it as in a book or paragraph, you're just breaking it down into symbol format. Whatever works man. Last ones look really cool.

PS. "JophASL" was just the previous poster with the Japanese woman Shintoism picture. I was just referring to the description style they used.

I think it's just more about the "highly detailed" "anatomically correct" "anatomy" etc descriptors. Also, the artstation ones naturally have more source to draw from since most of the subjects in there deal with very detailed and higher res faces. Just a thought.

the reason why i figured that out is because I wanted to see how accurately it can depict the human organs like Human heart. So that's why I even used that phrase in the first place. But it was slightly abstract even though very close

"My thing is I'm trying to synthesis faces that don't really exist, from scratch so to say. And as you can see there are limitations like details with eyes and all of that that im trying to over come with the equations and combinations as you can see descriptions only go so far."

Ah ok gotcha. Yeah I guess you're doing the same thing, but instead of describing it as in a book or paragraph, you're just breaking it down into symbol format. Whatever works man. Last ones look really cool.

PS. "JophASL" was just the previous poster with the Japanese woman Shintoism picture. I was just referring to the description style they used.

ahh gotcha yeah I want to get ready before i can get my hands on the dalle 2 thingy cause then I will implement all this things to take it to the limits. I'm pretty interested in 4d architecture and would love to tweak around and I have some awesome ideas

"I wanted to see how accurately it can depict the human organs like Human heart. So that's why I even used that phrase in the first place. But it was slightly abstract even though very close"

Yeah same. I was trying to get detailed pencil sketches like Da Vinci. I tried adding a few other things like "detailed schematic", "detailed blueprint" "medical illustration" and got some interesting results.

It does ears really well though! Granted the inner structure of the ear is already so unique per person so any odd shapes Dalle comes up with are very forgiving and look like they fit right in.

"I wanted to see how accurately it can depict the human organs like Human heart. So that's why I even used that phrase in the first place. But it was slightly abstract even though very close"

Yeah same. I was trying to get detailed pencil sketches like Da Vinci. I tried adding a few other things like "detailed schematic", "detailed blueprint" "medical illustration" and got some interesting results.

It does ears really well though! Granted the inner structure of the ear is already so unique per person so any odd shapes Dalle comes up with are very forgiving and look like they fit right in.

Wow this is incredible. I never thought of that actually. It just looks like a normal sketched ear. First time I can say it actually made something without those wierd issues that always come up. Please keep me update to what you found. I'm really curious

witness my latest abomination
if you wanna try it out and tweak:
Artstation style = highly detailed : ( subject =highly detailed💃🏻🧕🏼angelina jolie+(anatomically correct facial features + (highly detailed = 👩🏼‍🦰🧝🏼‍♀️)+((highly detailed and anatomically correct (realistic and highly detailed + anatomically correct and accurately shaped eyes=👁👁),highly detailed and anatomically correct👃🏼,highly detailed and anatomically correct👄)))