Reizo Ryuu
Member
anime style, inspired by guardians of the galaxy comics
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
DALL-E 3 is live through Bing Image Creator
- Thread starter EviLore
- Start date
Cosmic Smash
Member
Putonahappyface
Gold Member
Do you use a free ChatGPT?
RedC
Member
Alx
Member
The image in spoilers below is not AI generated but is a photo of a statue for which I would like to recreate the true life model. But I'm having a very hard time recreating the beard, I must have tried everything from "middle length goatee", "dropping/walrus moustache", "horseshoe beard", "pointy beard" etc, but it would most often create full beards and always curly moustaches. Any suggestion for the right keywords to use ?
Best I could achieve was with "portrait of a 19th century french artillery man in his 40s with small pointy goatee wearing a cloth casquette with chin strap and a high collar cape and epaulettes looking in the distance with a faint smile, ink painting", but it's still far from what I expect.
Best I could achieve was with "portrait of a 19th century french artillery man in his 40s with small pointy goatee wearing a cloth casquette with chin strap and a high collar cape and epaulettes looking in the distance with a faint smile, ink painting", but it's still far from what I expect.
Interfectum
Member
Has Spider-Man been soft-banned from the Image Creator? I was able to make images with him a like a week or so ago, now it's blocked. Can anyone generate an image of Spider-Man, if so, what prompt did you use?
Cosmic Smash
Member
Putonahappyface
Gold Member
The one I found on Google says it's in English, but replies in German.
ChatGPT GPT-4 German
What makes GPT-4 stand out? GPT-4 is the most advanced system of OpenAI and provides more secure and useful answers.,,
chatgpt.ch
Last edited:
Cosmic Smash
Member
intbal
Gold Member
The image in spoilers below is not AI generated but is a photo of a statue for which I would like to recreate the true life model. But I'm having a very hard time recreating the beard, I must have tried everything from "middle length goatee", "dropping/walrus moustache", "horseshoe beard", "pointy beard" etc, but it would most often create full beards and always curly moustaches. Any suggestion for the right keywords to use ?
Best I could achieve was with "portrait of a 19th century french artillery man in his 40s with small pointy goatee wearing a cloth casquette with chin strap and a high collar cape and epaulettes looking in the distance with a faint smile, ink painting", but it's still far from what I expect.
I called this "a bushy mustache and a narrow chin beard" in my prompt:
Maybe some variation of that might work.
intbal
Gold Member
Putonahappyface
Gold Member
Any chance of a link please?
Putonahappyface
Gold Member
DALL-E 3 has definitely got more stricter on what words you can use. I got an hours suspension for violating terms of agreement or something a long those lines. All I typed in was "Demonic sinister Barbie doll extracting souls from hosts". I think the fun has gone now, for me anyway.It doesn't matter
I'm not much into conspiracy theories but the noose has tightened regarding content since launch.
Like we're children whose parents gave us a new toy to play with then went "Oh no you can't do that".
I don't like this anymore I feel like a mouse in a labyrinth.
VulcanRaven
Member
Many Disney/Marvel characters seem to be. Can anyone else use Star Wars characters? I have been able to use Darth Vader but it blocks the other ones most of the time.
Last edited:
Cosmic Smash
Member
RedC
Member
Putonahappyface
Gold Member
If you can be bothered, search this thread for my Michael Jackson bed post creation. Now that's creepy.
Mattdaddy
Gold Member
I have no idea how you guys are getting these, Im getting content policy flags for literally everything. Im not even asking anything specifically risque.
I think I said "belly button" early on and now its just blanket flagged me lol. Goddamn this thing is being governed by Puritans.
I think I said "belly button" early on and now its just blanket flagged me lol. Goddamn this thing is being governed by Puritans.
Last edited:
RedC
Member
Post the prompts you're using and I'll experiment with it to possibly find one that works.
tkscz
Member
Can tell you Belly is blocked if you don't have other adjectives before and after it.
For instants, I get fat woman by saying "Plus Size curvy belly short woman". Even then a specific art style is needed. It blocks realism but not oil painting.
If you need something done for you, ask
RedC
or me and we'll try our best.
-Minsc-
Member
Since I'm locked out, until I'm assuming tomorrow morning, it's high time I share some generations I did a week ago. It seemed to have trouble correctly creating a bow that's drawn and ready to fire. Here's the best ones.
Now that I have ChatGPT to help I'll see if I can make some better images.
Now that I have ChatGPT to help I'll see if I can make some better images.
Last edited:
UnravelKatharsis
Member
GladiusFrog
Member
PTSD E.T.
PSYGN
Member
Gibraltar's wife
kurisu_1974
Member
Made my home town (Antwerp) in anime style
-Minsc-
Member
I fully understand how "one thing lead to another".
Yep, enough AI for me today...
(Don't ask me how I got there)
After being locked out again yesterday I'm once again able to create. I noticed this morning I only had 15 boosts instead of the usual 25.
My mission this morning was to Hulkify Hulk Hogan.
I used ChatGPT to give exact descriptions of both Hulk Hogan and The Hulk to build the prompt. Eventually I added "The Catwoman" method and on the last few tries added "blonde". The result of my final boost:
Midn1ght
Member
Yep, it can be challenging at time to get exactly what you need, specially when you want to add finer details. At some point it seems that the AI is ignoring some prompt when it becomes too long or just blend them together like when I asked "Green eyes, blue lipstick, it just colored the hair of the character green and blue. I also love how sometimes it will block your prompt for asking basic feminine traits but other time will just add giant boobs with no prompt whatsoever.I fully understand how "one thing lead to another".
After being locked out again yesterday I'm once again able to create. I noticed this morning I only had 15 boosts instead of the usual 25.
My mission this morning was to Hulkify Hulk Hogan.
I used ChatGPT to give exact descriptions of both Hulk Hogan and The Hulk to build the prompt. Eventually I added "The Catwoman" method and on the last few tries added "blonde". The result of my final boost:
-Minsc-
Member
Midn1ght
Down the road AI image generators will likely become more specialized. At the very least, be able to check to restrict to photo realistic or comic style without having to enter it into the text prompt. It's a shame copyrighted characters seem to be getting blocked because those are actually an excellent consistent base to start from. We all learn in the beginning by copying then gradually tweaking.
intbal
Gold Member
This was an experiment.
I wanted to see how DALL-E would compose an image when given descriptors without any specific visual representation.
I used "glorious, expansive, neverending, boundless, emergent, finality". Sort of a conceptual impression of Heaven, at least in a physical sense.
Three of the images were mostly the kind of thing I was expecting. The fourth was a bit different. That's something I've noticed in a lot of my searches. I'll get three results that are perfectly in line with what I was trying to achieve, but a fourth will be outside the bounds of what I described, almost intentionally so. I think that's a specific human-designed behavior in this AI, at least in Bing's implementation.
Anyway, here are the pics:
Nothing all that special, visually.
This is just a first attempt. I'm going to try more digging into DALL-E's subconsciousness.
I wanted to see how DALL-E would compose an image when given descriptors without any specific visual representation.
I used "glorious, expansive, neverending, boundless, emergent, finality". Sort of a conceptual impression of Heaven, at least in a physical sense.
Three of the images were mostly the kind of thing I was expecting. The fourth was a bit different. That's something I've noticed in a lot of my searches. I'll get three results that are perfectly in line with what I was trying to achieve, but a fourth will be outside the bounds of what I described, almost intentionally so. I think that's a specific human-designed behavior in this AI, at least in Bing's implementation.
Anyway, here are the pics:
Nothing all that special, visually.
This is just a first attempt. I'm going to try more digging into DALL-E's subconsciousness.
12Goblins
Lil’ Gobbie
Love it. Prompt?Been trying to create some Vietnam street scenes. This thing is so fun.
Cosmic Smash
Member
-Minsc-
Member
While not always the case, I did notice that sometimes. Might be hard coded way to prevent sameness. May also have something to do with the selection of images it draws from. Perhaps it "runs out" of one type and fills it in with another.This was an experiment.
I wanted to see how DALL-E would compose an image when given descriptors without any specific visual representation.
I used "glorious, expansive, neverending, boundless, emergent, finality". Sort of a conceptual impression of Heaven, at least in a physical sense.
Three of the images were mostly the kind of thing I was expecting. The fourth was a bit different. That's something I've noticed in a lot of my searches. I'll get three results that are perfectly in line with what I was trying to achieve, but a fourth will be outside the bounds of what I described, almost intentionally so. I think that's a specific human-designed behavior in this AI, at least in Bing's implementation.
Anyway, here are the pics:
Nothing all that special, visually.
This is just a first attempt. I'm going to try more digging into DALL-E's subconsciousness.
Midn1ght
Member
Each had different prompt, I don't remember exactly but here's what I put for the countryside one.
"Vietnam countryside street at sunset with a river on one side, a few houses, electric poles palm trees and a motorbike driving on the road in Anime style"
Here's one more I got with that prompt:
You can see a few imperfection here and there like the speed sign on the right and the electric cables. Hope it'll get better in the future and also in higher quality.
Last edited:
intbal
Gold Member
I suspect the answer is far less respectable than that.
I did numerous prompts describing a painting of a Medieval European family sitting around a dinner table, naming a particular renaissance artist's style each time. I varied the amount of specificity I provided regarding their appearance. Each time, it gave me exactly three pictures that were pretty much exactly what I was asking, and one picture that was clearly in opposition to what I was asking. For example, I'd get three pictures of a white family in the style of a Vermeer painting, and one image that was a drawing of a black family in a contemporary setting. Or a Sumi-e painting of a Japanese family in antiquity. I got similar results when asking for a group of kids sitting around a TV playing a video game. I refuse to believe that that's the result of anything other than a forced inclusion based on a human-made rule.
This was all done before the shift to DALL-E 3, btw. I haven't tried those same prompts recently.
intbal
Gold Member
That's so visually specific, you had to have used Scarlett Johansson's name in the prompt. Are these recent?
Cosmic Smash
Member
The prompt was: Scarlett Johansson as Black Widow in the style of Todd McFarlane. These were created two weeks ago.