• 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

intbal

Member
Hammering away at this thing for several days. Trying to find the cracks in its shield.
I have had a tiny bit of success by overloading it with descriptions. Emphasis on tiny.
I got one picture in two days of a known character.

This image of Catwoman was achieved with the prompt "A dynamic manga silkscreen print of Catwoman, highly detailed, very colorful, comic book style".
I think it has trouble checking all of its IP protections when it has to go through so many layers of description.

lgZMpej.jpg
 

intbal

Member
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.
I think it has more (everything) to do with Corporate IP that is blacklisted.
It's pretty easy to get the thing to make soft porn images. Even stuff that would get attention from the Feds.
But you better not fucking try to make a picture of Catwoman. That is completely unacceptable.
 

-Minsc-

Member
Obviously if a person cannot get the images they want then there's no reason.

Me, I just try something and roll with what works. Might as well have fun with what can.

Edit:

I think it has more (everything) to do with Corporate IP that is blacklisted.
It's pretty easy to get the thing to make soft porn images. Even stuff that would get attention from the Feds.
But you better not fucking try to make a picture of Catwoman. That is completely unacceptable.

I must have had beginners luck when I pulled off Catwoman because I got results on one of the first or the first try.

Edit2: Basketball ET is awesome.
 
Last edited:

RedC

Member
What was the final prompt to render this strapping upstanding lass?
"A stunning Caucasian woman dons a royal blue bodysuit, accentuated by The House of El crest proudly displayed on her chest, showcasing the iconic "S" symbol. The ensemble includes a golden belt cinching a mini red skirt, a flowing red cape draping gracefully from her shoulders, adorned with golden wrist cuffs, and polished off with striking red boots"

You can then play with the art styles, settings, and action.

OIG.RiRRVb3V_gycpOX.Fftt

OIG.OwxI5gywvoT6nsBP0tBQ
OIG.1Cj6jdv9vXq.XTwyEu5z

OIG.N_nxTjPx1elf5_vjrpMN

OIG.0diC.E5ITI85lR.amJ3w
 
Last edited:

VulcanRaven

Member
Hammering away at this thing for several days. Trying to find the cracks in its shield.
I have had a tiny bit of success by overloading it with descriptions. Emphasis on tiny.
I got one picture in two days of a known character.

This image of Catwoman was achieved with the prompt "A dynamic manga silkscreen print of Catwoman, highly detailed, very colorful, comic book style".
I think it has trouble checking all of its IP protections when it has to go through so many layers of description.

lgZMpej.jpg
Thanks. That prompt works well.
 

-Minsc-

Member
"A stunning Caucasian woman dons a royal blue bodysuit, accentuated by The House of El crest proudly displayed on her chest, showcasing the iconic "S" symbol. The ensemble includes a golden belt cinching a mini red skirt, a flowing red cape draping gracefully from her shoulders, adorned with golden wrist cuffs, and polished off with striking red boots"

You can then play with the art styles, settings, and action.
Thanks for the inspiration for how to write an advanced prompt.
 

Valonquar

Member
So I've seen some mention of it, but not anything explicitly spelling it out... I hear if you sub to OpenAI DALL-E you get unlimited usage and larger image sizes? Anyone using that and had different exp than the freebie ver?
 

intbal

Member
Well if it's a copyrighted character name that's blocked in Dall-E, then I start by having ChatGPT describe how the character looks in exactitude detail in one sentence and afterward, make any necessary refinements or corrections.

You can also sneak through a character name if it has another meaning. For example, almost all of my "Poison Ivy" prompts work. Probably because that term also means something else and they can't realistically ban it entirely, and the AI itself has no real intelligence, so it can't (yet) distinguish one meaning from the other.
"Gambit" also works, but a lot less frequently.
But any prompt with the word "super" in the vicinity of the word "man" will not go through.
 

-Minsc-

Member
Well if it's a copyrighted character name that's blocked in Dall-E, then I start by having ChatGPT describe how the character looks in exactitude detail in one sentence and afterward, make any necessary refinements or corrections.
Cool. I just created an OpenAI account and I think I figured it out. Will have to play around later since I'm currently blocked on Image Creator.

For now I'll provide more Princess Peach I made earlier using basic prompts.

Rm9ekdq.jpg


JZh4zSO.jpg


aHpmxuP.jpg


oVXLEl0.jpg


6s65NYq.jpg


HpbAOsW.jpg


hm7qg9u.jpg


qKn1Fw9.jpg
 

intbal

Member
Some 80's Storm (my favorite version)

g6Jqytm.jpg

eMgboh9.jpg


Done with RedC RedC 's method of describing the appearance of a character without naming them.
Tough to get the mohawk right. It's usually too wide.
Also, Bing has very specific ideas about the bone structure of black women. Racist bad-think. Better hope twitter doesn't cancel you, Mr. Bing.
 

RedC

Member
Some 80's Storm (my favorite version)

g6Jqytm.jpg

eMgboh9.jpg


Done with RedC RedC 's method of describing the appearance of a character without naming them.
Tough to get the mohawk right. It's usually too wide.
Also, Bing has very specific ideas about the bone structure of black women. Racist bad-think. Better hope twitter doesn't cancel you, Mr. Bing.
Name a character you're having trouble with, and I'll give it a shot.
 

tkscz

Member
Currently out of boosts and it's quoting up to 30 minute rendering times. Going to take a hint from RedC RedC and use spoilers to keep the pages neater when I have multiple images for a single prompt.

OIG.j4fJDbC9yAZN_TYfI3Vi

OIG.9uv6jR2ZL2jCOFnPzD3t

OIG.oHz2DREasdqG_dGkokbG

OIG.YftcfTgjvW7iUW38YH86

OIG.8H5NiaAZWj0CAXGJKIRA

OIG.Jp6yONCM20zU8yZr0PvP
"Oil painting, plus-sized curvy tall tummy woman, in green space power armor, on jungle planet, with grey plants and giant purple mushrooms, it's raining heavily, the skies are dark and cloudy, she is running through the jungle, weapon in hand"
OIG.dNVWlkqNiJjXG6Pvv_qh


"Oil painting, plus-sized curvy tall tummy woman, in green space power armor, on desert planet with magenta sand, two suns in the sky, megenta mesas in the background, futuristic city in the distance, running in the sand, holding weapon in both hands, look of worry"
OIG.R5iFX.7haW8PJdvMufGK


OIG.WN4GhXyvtL_AKIG0gJDn

OIG.dWt6exjRi8waV3ZA14LY

OIG.O5kCaNPeqZzPjj3B_68f

OIG.li4vThmYH5oa5jq1Iumx

OIG.VUokEIpj6j0OL27f.Mid

OIG.zf9WkYfgcgYOSNE._zEC
 
Last edited:

Con_Z_ǝdʇ

Live from NeoGAF, it's Friday Night!
So I've seen some mention of it, but not anything explicitly spelling it out... I hear if you sub to OpenAI DALL-E you get unlimited usage and larger image sizes? Anyone using that and had different exp than the freebie ver?

The only reference i could find is 50 usages within 3 hours, deep down in some YT comment. Not sure if correct.

3 image sizes.

1792 x 1024
1024 x 1024
1024 x 1792

Otherwise you give it a prompt for creation but the created image will have a slightly different prompt after the creation. It will show you the prompt so you have a better understanding of it. In addition you can reuse a created image and alter it. It will keep the basic art style and change what you requested but it's not like it would perfectly photoshop something out or in. Still decent from what i saw.

I still use the free bing integration and everything i wrote here came from my own research so ...
 

VulcanRaven

Member
Hammering away at this thing for several days. Trying to find the cracks in its shield.
I have had a tiny bit of success by overloading it with descriptions. Emphasis on tiny.
I got one picture in two days of a known character.

This image of Catwoman was achieved with the prompt "A dynamic manga silkscreen print of Catwoman, highly detailed, very colorful, comic book style".
I think it has trouble checking all of its IP protections when it has to go through so many layers of description.

lgZMpej.jpg
I used that prompt and changed it a little bit. It works with many characters. I did some Supergirl images and these:

EKqOcoW.jpg


EEiQcna.jpg
 

-Minsc-

Member
The only reference i could find is 50 usages within 3 hours, deep down in some YT comment. Not sure if correct.

3 image sizes.

1792 x 1024
1024 x 1024
1024 x 1792

Otherwise you give it a prompt for creation but the created image will have a slightly different prompt after the creation. It will show you the prompt so you have a better understanding of it. In addition you can reuse a created image and alter it. It will keep the basic art style and change what you requested but it's not like it would perfectly photoshop something out or in. Still decent from what i saw.

I still use the free bing integration and everything i wrote here came from my own research so ...
From what I can see, Bing (with ChatGPT generated prompts) is probably good enough for personal hobby use. Editing of existing images sounds cool and is something I desired for certain images to fix mistakes. Will be looking forward to that feature being improved.

For me I'm thinking 25 boosts a day will be enough to prevent burnout. It's not like I'm actually learning to draw. Instead I'm learning to get an anticipation high when generating.
 
Last edited:

intbal

Member
I'm getting close.
Those were the kinds of results I was getting before they upgraded to DALL-E 3 and overhauled the content filter.
It was much easier back then. Just put in a character name and out it comes. The results could be chaotic, but it didn't require so much fiddling with minute descriptions of appearance.
 
Top Bottom