Post your AI images
Discussion
A picture tells a thousand words.
And this highlights the issues with NN AI and LLM.
The language model needs a thousand words to make a single picture, but the image descriptions or references aren’t as absolute as a language.
Images haven’t been described with precision, and images are abstract.
One persons stark might be another’s beautiful.
The problem for me with all these models is, the more words you add the less control or accuracy you seem to get.
I’m also of the view that in many scopes everything is tending to look the same.
When I look through most of these systems outputs nothing stands out.
Either prompts aren’t specific enough, or the prompting isn’t good enough to capture creative or explicit ideas.
Most of the time you get better results just searching Google for an image with the same prompt.
And right now given copyright concerns, accuracy concerns, demonic hands, eyes, teeth, yadda yadda, right now this is most useful as a base to work off… and for that any old existing image works fine too.
Ie, why pay for AI to make reference images when Google will find them for you to start with?
And this highlights the issues with NN AI and LLM.
The language model needs a thousand words to make a single picture, but the image descriptions or references aren’t as absolute as a language.
Images haven’t been described with precision, and images are abstract.
One persons stark might be another’s beautiful.
The problem for me with all these models is, the more words you add the less control or accuracy you seem to get.
I’m also of the view that in many scopes everything is tending to look the same.
When I look through most of these systems outputs nothing stands out.
Either prompts aren’t specific enough, or the prompting isn’t good enough to capture creative or explicit ideas.
Most of the time you get better results just searching Google for an image with the same prompt.
And right now given copyright concerns, accuracy concerns, demonic hands, eyes, teeth, yadda yadda, right now this is most useful as a base to work off… and for that any old existing image works fine too.
Ie, why pay for AI to make reference images when Google will find them for you to start with?
Edited by Mr Whippy on Friday 15th March 12:23
After today's news about Tesco losing their appeal to Lidl over the Clubcard logo looking like the Lidl logo because it's a yellow circle in a blue square, here's a completely different Clubcard logo as a blue circle in a yellow square. There's no way Lidl could complain about this
"yellow background. a dark blue solid colour circle with a red outline. the words "TESCO Clubcard" in the centre, white text"
"yellow background. a dark blue solid colour circle with a red outline. the words "TESCO Clubcard" in the centre, white text"
ajprice said:
After today's news about Tesco losing their appeal to Lidl over the Clubcard logo looking like the Lidl logo because it's a yellow circle in a blue square, here's a completely different Clubcard logo as a blue circle in a yellow square. There's no way Lidl could complain about this
"yellow background. a dark blue solid colour circle with a red outline. the words "TESCO Clubcard" in the centre, white text"
I suggest you present that to the Tesco Board, with an invoice for..... A MILLION POUNDS! "yellow background. a dark blue solid colour circle with a red outline. the words "TESCO Clubcard" in the centre, white text"
Hedobot said:
White text ?
Yeah, there were others without the red outline, and a lot with misspellings, that was one of the good ones.This one with the same prompt... wasn't
Bing definitely knows the Tesco logo though
Edited by ajprice on Wednesday 20th March 17:02
Edited by ajprice on Wednesday 20th March 19:24
I’ve used AI for this kinda thing… just create pages of ideas and pick a few out, refine with further variants, then work-up manually.
BUT, this was always possible just going on stock logo/image sites, picking out stuff that gave you ideas, work up manually.
So a bit of saved time, maybe, but it’s not doing the job for you yet… and after a fair few years it’s as ignorant as ever.
Johnny Five painted pink, dancing on top of freedom tower in New York at sunset
As with many prompts for something like this, it thinks you want to see what you’ve written.
Ie, mention being on the moon, and the moon is in the background.
Freedom tower in this case gets you freedom tower.
And that johnny five is rubbish
Sheep made from carrots and cauliflower
Amazing.
Now, as we all know, a client will say yes perfect, but…
And that but, that fine tweak, is now impossible to achieve. Let’s say they want one more sheep.
Or parsnips not carrots, and then it creates a totally different image.
AI gives you something, a very nice novel something, but to pretend you’ve got any control is a fallacy for now.
For people who are happy with superficiality, or quality over control, great.
But then nothing ‘proper’ really fits those features. I never get a client who just wants eye-candy, as they can already find eye-candy on stock sites that just looks good but doesn’t fit a specific need.
Amazing.
Now, as we all know, a client will say yes perfect, but…
And that but, that fine tweak, is now impossible to achieve. Let’s say they want one more sheep.
Or parsnips not carrots, and then it creates a totally different image.
AI gives you something, a very nice novel something, but to pretend you’ve got any control is a fallacy for now.
For people who are happy with superficiality, or quality over control, great.
But then nothing ‘proper’ really fits those features. I never get a client who just wants eye-candy, as they can already find eye-candy on stock sites that just looks good but doesn’t fit a specific need.
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff