Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

Reality

15 (68.2%)

Universe

4 (18.2%)

The World

3 (13.6%)

Total Members Voted: 22

[Imp]

It's certainly not unsurmountable

Well... Better then what that guy showed, but still not actually working. Doesn't seem to be surmounted.

A possible key for images like this, is to consider where in reality or fantasy (whatever the models might have been trained on images of) that we see totally full glasses.

Some Japanese style tea ceremonies, and expresso shots sometimes are like this, so I explored there.  You can also use related specific terms, like 'Mǎn bēi' to indicate the appropriate fullness with the Japanese style tea.

Spoiler (click to show/hide)

Spoiler (click to show/hide)

Spoiler (click to show/hide)

[Starver]

"Eliza, ask Parry to tell Claude to use ChatGPT to get Perplexity to ask Gemini to request Grok to have Watsonx prompt Stable Diffusion to present AlphaZero with a puzzle to..."

(I'm sure we can get them to tie themselves in knots. Even if we need to ask an AI how to do it!)

[King Zultan]

Maybe one day we'll get an AI that knows to create hands, people eating noodles, full glasses, and oil paintings that actually look normal.

Or the AIs will just train us to stop asking for those things...

Happened to notice this post, wanted to play too.

I'm pretty comfortable with AI.  No, I don't think they're perfect, and I don't care if they are or not, more fun for me if they're not because then I have more to do.

Spoiler: But here's a prompt from me to ChatGPT about this hands, people eating noodles, full glasses, and oil paintings thing.  Did I totally miss the point? (click to show/hide)

Spoiler: ChatGPT's prompt to Dall-E (click to show/hide)

Seems like it's gotten better at some things but it still doesn't seem to understand the look of an oil painting, maybe not including hyperealistic would make it work better.

[brewer bob]

Spoiler (click to show/hide)

AI fingers strikes again

[martinuzz]

I'd say it would be easy by now to train an AI to draw hands in all kinds of positions, with the correct number of digits. Just plug VR sets that use virtual hands into them and feed them endless data of the camera input of people's hands manipulating objects in the virtual world

[Criptfeind]

A possible key for images like this, is to consider where in reality or fantasy (whatever the models might have been trained on images of) that we see totally full glasses.

Some Japanese style tea ceremonies, and expresso shots sometimes are like this, so I explored there.  You can also use related specific terms, like 'Mǎn bēi' to indicate the appropriate fullness with the Japanese style tea.

I dunno if I can accept the idea that the key to solving the inability to show a full glass of wine problem is to give up and ask the AI for something else.

[Naturegirl1999]

PTW

[Loud Whispers]

I dunno if I can accept the idea that the key to solving the inability to show a full glass of wine problem is to give up and ask the AI for something else.

Me asking the AI bartender for a full glass of wine in the style of Japanese tea culture