Spaces:

dalle-mini

/

dalle-mini

like 5.43k

Running

hahahahahLMAO

Try Salvador Dali style with high level of detail: Subject= woman👩, subject detail: anatomically accurate real blue eyes=detailed👁👁and Blue earrings

Prompt: Salvador Dali style with high level of detail: Subject= woman👩, subject detail: anatomically accurate real blue eyes=detailed👁👁 cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve + Blue sapphire earrings

Prompt: Photograph portrait of a old woman, studio lighting, highly detailed, anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve

WE ARE ALMOST THERE niiicee

given there must be hundreds, thousands of faces in the training data what are your thoughts about the difficulty of reproducing faces?

Prompt: (Photograph, picture, photo) (medium close-up, medium closeup) (portrait photo) of an (old woman crying, tears, sadness, unhappy, pain), (studio lighting, photographic studio lighting) (fine detail, highly detailed, finely detailed) (anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

Trying ThunderJames' intuition of 'equation style' prompts

Prompt: (Photograph, picture, photo) (wide shot, full body shot, 24mm lens, 30mm lens) (portrait photo) of an (old woman sitting in a chair, old woman sitting down) (old woman crying, tears, sadness, unhappy, pain), (studio lighting, photographic studio lighting) (fine detail, highly detailed, finely detailed) (anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

Comment: The lens specification looks promising. The non-verbal emotion in the posing is outstanding. But why has she turned into a furry? :)

given there must be hundreds, thousands of faces in the training data what are your thoughts about the difficulty of reproducing faces?

it really depends i guess , either you want to black box it and just let the AI figure it out. Or we figure out a better impute language with a higher level of accuracy. Thats why I'm trying to figure out how I can specify toward certain detail and how I can split the process while the AI still understands. Thats why I started simple with basic features like subject ground and background. But I believe there are far more spesification's that are yet to be discovered. I don't know jack shit about programming so for me its all just intuitive and simple trial and error.

Prompt: (Photograph, picture, photo) (wide shot, full body shot, 24mm lens, 30mm lens) (portrait photo) (studio lighting, photographic studio lighting) of an (old woman sitting in a chair, old woman sitting down) (old woman's face is crying, the old woman has tears in her eyes, sadness, unhappy, pain) (anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

Comment: Nope!

"either you want to black box it and just let the AI figure it out. Or we figure out a better impute language with a higher level of accuracy"

Yea. I like the idea of some level of control

Tried something similar (first try with the equation) but no eyes at all

Prompt: (Photograph,still,frame)(full body shot) of a (little girl, young girl) with a candy (crying, sad, tears) (anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

Prompt: (Photograph, picture, photo) (wide shot, full body shot, 24mm lens, 30mm lens) (portrait photo) of an (old woman sitting in a chair, old woman sitting down) (old woman crying, tears, sadness, unhappy, pain), (studio lighting, photographic studio lighting) (fine detail, highly detailed, finely detailed) (anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

Comment: The lens specification looks promising. The non-verbal emotion in the posing is outstanding. But why has she turned into a furry? :)

for example here you can see how the AI just gives up on the faces. Thats why I sometime "remind" it with emojis but also stuff like
subject and subject detail
here for example I might try out

(Photograph, picture, photo) (wide shot, full body shot, 24mm lens, 30mm lens) (portrait photo): Subject= grandma, Subject details: (Pose = sitting in a chair, sitting down ,)+ (Emotion=crying, tears, sadness, unhappy, pain) + (face=anatomically correct human eyes, detailed: cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve)

I then like to watch the results and see where the input and the output don't connect. Then I zero in on that zone and just work it out

yeah i ran into the same issue , Im currently working on it

this is why you saw my equation getting complicated, because I was tweaking to combat this issue. In dalle mini - verbal input alone is not enough

A lack of detail in input can result to a lack of detail in output I seem to observe

@ThunderJames "Then I zero in on that zone and just work it out" I do something similar. I went back to one up top that looked better (no equation)

Prompt: detailed photograph of woman's face smiling, blue eyes, cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve

Comment: Gotta do some chores. I expect y'all will have this all worked out when I get back! :)

Gents, so I see everyone is headed in a similar direction so let's make this badass and do it right. And especially knock it out with all of the server busy issues.

I'm setting up some servers with a couple T4's and RTX 90's as GPUs and getting a few devs to build & customize the code to make this even more amazing.

Let's see what we can come up with - see you all on the other side...
https://discord.gg/2DH2Cxg99S

...

@ThunderJames "Then I zero in on that zone and just work it out" I do something similar. I went back to one up top that looked better (no equation)

Prompt: detailed photograph of woman's face smiling, blue eyes, cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve

Comment: Gotta do some chores. I expect y'all will have this all worked out when I get back! :)

im gonna do my best lol

@ThunderJames "Then I zero in on that zone and just work it out" I do something similar. I went back to one up top that looked better (no equation)

Prompt: detailed photograph of woman's face smiling, blue eyes, cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve

Comment: Gotta do some chores. I expect y'all will have this all worked out when I get back! :)

yeah the issues arise when you want more information of things to come together as one. then the inputs kind of become challenging , I don't know if its too vague or something

Prompt: symmetrical human face photograph
Comment: Chores can wait. I think Dalle-mini struggles with facial symmetry so I added "symmetrical "

Prompt: photograph of human face || Facial bilateral symmetry || anatomically correct human eyes, cornea, iris, pupil, aqueous humor, lens, vitreous humor, retina, and optic nerve||
Comment: tried symmetrical again. used pipes || but I don't think it helps. I feel the many words describing the eye makes Dalle-mini focus on that idea. Would be nice to work out how to 'weight' different ideas that are joined together in a single prompt.

Prompt: symmetrical human face photograph
Comment: Chores can wait. I think Dalle-mini struggles with facial symmetry so I added "symmetrical "

Jack you need to get past the censoring for that you must find a way for the computer to understand your other inputs in combination with the eye prompt you made as one. It needs to interpret it as that information belonging together. Right now it only sees them as two different things gets why you get either the one eye or the blurry or not even really visible eyes with the face.
I'm still working on that part