Support NeoGAF

[Alx]

Funny that popular memes are probably so over-trained in the neural network that you can recreate an almost identical picture based on its description.

Prompt :
"photorealistic scene in the street : a young woman in a red dress walking towards the camera in the first plane, a man in a blue flannel shirt and his girlfriend in a blue dress going in a different direction. The man turns his head with a whistling expression to check the first woman, his girlfriend looks at him with a shocked and angry face."

Result :

Reference :

It even reused the same background buildings and passers by, and "fixed" the girlfriend clothes since she actually wasn't wearing a dress.

 

[-Minsc-]

Funny that popular memes are probably so over-trained in the neural network that you can recreate an almost identical picture based on its description.

Prompt :
"photorealistic scene in the street : a young woman in a red dress walking towards the camera in the first plane, a man in a blue flannel shirt and his girlfriend in a blue dress going in a different direction. The man turns his head with a whistling expression to check the first woman, his girlfriend looks at him with a shocked and angry face."

Result :

Reference :

It even reused the same background buildings and passers by, and "fixed" the girlfriend clothes since she actually wasn't wearing a dress.

Fake news is gonna be great.

 

[Dr.D00p]

Potential videogame artists...find a new career.

 

[Wildebeest]

Funny that popular memes are probably so over-trained in the neural network that you can recreate an almost identical picture based on its description.

It is called overfitting. When the model fits the sample data so tightly that it basically reproduces it. There was a fuss when Stable Diffusion launched that it was reproducing watermarks for copyright images that they didn't have the right to use.

 

[-Minsc-]

Potential videogame artists...find a new career.

Does the upscaling only work so well because there's plenty of higher resolution reference to draw from? There's also plenty of idyllic images of brunettes to draw from. Basically, is it doing exactly what happened below?

Funny that popular memes are probably so over-trained in the neural network that you can recreate an almost identical picture based on its description.

Prompt :
"photorealistic scene in the street : a young woman in a red dress walking towards the camera in the first plane, a man in a blue flannel shirt and his girlfriend in a blue dress going in a different direction. The man turns his head with a whistling expression to check the first woman, his girlfriend looks at him with a shocked and angry face."

Result :

Reference :

It even reused the same background buildings and passers by, and "fixed" the girlfriend clothes since she actually wasn't wearing a dress.

I wonder, at its current level, how well does AI work creating cohesive original characters?

My questions aside, AI definitely will make things like remastering much simpler.

 

[Alex Anderson]

Okay the well is running a little dry on this. And man it's creating slow this morning.

 

[Croatoan]

Shout out to Croatoan

I used Taylor Swift as the muse.

Spoiler: More

Very, very good. Did you edit them? Btw, how are people getting celebrities without getting mega banned from bing? Is the a specific prompt? Actually would love to know your prompt in general as those are some very detailed images.

 

[ÆMNE22A!C]

Can someone make me

 

[Croatoan]

So I have come across a prompt that is spitting out some straight fire. Here it is:

(Optional Decade) (genre) artwork with a (main theme) theme blended with (additional theme 1) and (additional theme 2), in the style of heavy metal magazine covers, (subject and content), (additional keywords)

For example:

1980s science fiction artwork with a dystopian theme blended with gothic fantasy and horror, in the style of heavy metal magazine covers, a red headed space woman wearing leather shorts, muted colors, dramatic

Some other examples with different prompt variables plugged in:

Spoiler: More images

 

[RedC]

Funny that popular memes are probably so over-trained in the neural network that you can recreate an almost identical picture based on its description.

Prompt :
"photorealistic scene in the street : a young woman in a red dress walking towards the camera in the first plane, a man in a blue flannel shirt and his girlfriend in a blue dress going in a different direction. The man turns his head with a whistling expression to check the first woman, his girlfriend looks at him with a shocked and angry face."

Result :

Reference :

It even reused the same background buildings and passers by, and "fixed" the girlfriend clothes since she actually wasn't wearing a dress.

 

[RedC]

So I have come across a prompt that is spitting out some straight fire. Here it is:

(Optional Decade) (genre) artwork with a (main theme) theme blended with (additional theme 1) and (additional theme 2), in the style of heavy metal magazine covers, (subject and content), (additional keywords)

For example:

1980s science fiction artwork with a dystopian theme blended with gothic fantasy and horror, in the style of heavy metal magazine covers, a red headed space woman wearing leather shorts, muted colors, dramatic

Some other examples with different prompt variables plugged in:

Spoiler: More images

 

[Croatoan]

Also I found if you change the prompt above slightly you can get some great character sheets or turnarounds if you are trying to come up with concept art or just neat ideas.

Example: 1980s horror with a Christmas theme blended with gothic and fantasy, in the style of heavy metal magazine covers, a character concept sheet of a brunette female survivor wearing a red dress, muted colors, dramatic

Other examples with different variables:

Add using a similar prompt:

 

[RedC]

 

[RedC]

 

[Alex Anderson]

 

[Wildebeest]

Can someone make me

 

[ÆMNE22A!C]

Thanks Wildebeest !!

 

[Alx]

This time I tried to replicate the « evil little girl » meme but it kind of mixed it up with another one.

 

[RedC]

This time I tried to replicate the « evil little girl » meme but it kind of mixed it up with another one.

 

[Alx]

Nice. The little girl is somewhat original, but again the background is clearly extracted from the source material.

 

[RedC]

Nice. The little girl is somewhat original, but again the background is clearly extracted from the source material.

The prompt I used was:

"a little girl with a mischievous smirk stands in front of a burning house, her expression adding an ironic and humorous twist to the seemingly calamitous scene"

You can experiment by changing things in that prompt and seeing what it generates.

I added black for a little black girl in the prompt and this was the result:

 

[Alex Anderson]

 

[VulcanRaven]

 

[RedC]

Very difficult to generate an image but I used Kim Kardashian as the base model.

 

[ADiTAR]

Thanks to the prompts Croatoan I tried to create the NCU.

 

[Radioskugga]

 

[Radioskugga]

 

[Lukaku's First Touch]

Anyone know where I can upload an image myself and have an AI make changes to it based on prompts? The ones I tried were garbage.

 

[Alex Anderson]

 

[Badlucktroll]

I had it show me what the blair witch looked like in the scene they planned in the original movie that was never shot where it was supposed to be a woman with long black hair staring at the main characters

 

[Alex Anderson]

Going for one of those cheesy 1980s multilayered portraits, lol

Trippy

 

[Solarstrike]

 

[ADiTAR]

NCU, now in 7th century oil paintings style.

 

[tkscz]

 

[ChuckeRearmed]

I am curious what the next generation of Dall E will be available to achieve. I guess they issue with fingers and eyes might be resolved.

 

[ADiTAR]

Different styles collide.

 

[RedC]

Spoiler: More

Spoiler: More

Spoiler: More

 

[Radioskugga]

 

[Alex Anderson]

Pfft. It won't let me make Abe Lincoln dunking a basketball. Or even "bearded president with a tophat"

Edit

There we go

 

[DeafTourette]

 

[ADiTAR]

Mind sharing your prompts? even in private msg? I'm just getting into it.

 

[RedC]

 

[DeafTourette]

Mind sharing your prompts? even in private msg? I'm just getting into it.

For the woman martial artist, I typed:

"with Nathalie Emmanuel as the base, create a woman using martial arts in a Chinese kung fu movie"

For the Avatar one, I typed: "create scenes where characters from Avatar The Last Airbender are fighting the Fire Nation "

Basically just type what you want to see.

 

[RedC]

 

[ADiTAR]

How do you get past the likeness? it keeps telling me it violates their T&C. When I say something like, based on Taylor Swift or whatever.

 

[Radioskugga]

 

[ADiTAR]

Happy Hannukah!

 

[Alex Anderson]

The last image makes me want my gamefly rental of kirby epic yarn 3ds to arrive right now!

 

[Alex Anderson]

I just asked for a woman surfing in oahu *shrug*

 

[Insane Metal]

Generated by Bing Image Creator

Boost your imagination, create your own AI images with Bing Image Creator!

www.bing.com