Heh, played around a bit with ChatGPT. Neat little tool, but I don't see how people actually use this for anything serious like novel writing, given it's really shit at remembering details, which really would be of the few things I would actually expect an AI to be really good at.
Artwork Messing around with AI, A Art thread.
- Thread starter Thegoldenboy2188
- Start date
Carrot of Truth
War is Peace
I have to admit this is the aesthetic I have been searching for, Basically its a "Cyber Gothic" aesthetic where you still have a high tech civilization that is still technologically progressing but culturally is like 15th century Europe.
On the whole I do not like AI.
I tried it. I got a very detailed image, far too detailed based on the simple description I gave. Therefore AI must be using stored images in its memory banks. I saw a picture of a steampunk anthro-vixen, she looked like a copper-skinned human woman with the shaggy head and tail of a vixen. It had to be a combination of several different images. Who did them? Can he claim any rights if his image(s) were used?
Much of it has the same look. You do not have the variety you get when people develop their own styles. Also, can AI learn? If someone wants to have a picture of Krystal the blue vixen can AI look it up on the Internet? If so then which images will it use? Images that are the property of someone else? Can anyone using AI claim the art as his own? I've seen terrific images but are they those of the artist?
I tried it. I got a very detailed image, far too detailed based on the simple description I gave. Therefore AI must be using stored images in its memory banks. I saw a picture of a steampunk anthro-vixen, she looked like a copper-skinned human woman with the shaggy head and tail of a vixen. It had to be a combination of several different images. Who did them? Can he claim any rights if his image(s) were used?
Much of it has the same look. You do not have the variety you get when people develop their own styles. Also, can AI learn? If someone wants to have a picture of Krystal the blue vixen can AI look it up on the Internet? If so then which images will it use? Images that are the property of someone else? Can anyone using AI claim the art as his own? I've seen terrific images but are they those of the artist?
Yes, that's literally how AI works. It's nothing BUT learning. It's basically how the human brain operates but on a much, much simpler level.
Nope, you have to go out of your way to feed it manually. You wouldn't want an AI to train off of unmoderated datasets anyways because a picture of a cucumber could be labelled "Krystal Fox" and all of a sudden your pictures are green and elongated.
Whatever pictures you want it to use.
I mean Krystal Fox is an owned IP, the vast majority of Krystal Fox images are in fact copyright infringement to begin with, and 'learning via looking at' is not copyright infringement in and of itself. You and I could draw Krystal Fox with some practice but the mere ability to do so is not illegal....Ehrm....Yet at least, that seems in the interests of big corporations.
Depends on the country but it's possible yeah, and frankly it's going to get a lot more common.
AI art images are belonging to whoever prompted them as much as a photograph belongs to a photographer who set up all the lenses, shutter speeds, angles and color corrections.
Yeah in theory an AI art generator can do that shit automatically, so can a camera. But both requires someone to actually set it all up in the first place and that's the artist.
TheRejectionist
TheRejectionist
As someone who likes to write timelines and other stuff AI art things can be useful as a placeholders when you can't hire artists themselves but I would never claim them as mine.
Following up on this, AI LLMs like ChatGPT and GROK have something I'd describe as drop-off points. Because of the immense power requirements to run them, internal memory is severely limited, but w/regards to scope as well as duration of any kind of "conversation". After a certain threshold has been crossed, everything from early on in your conversation with the LLM will drop off a cliff, completely vanishing from the LLM's understanding and context of your convo. Which then leads to... interesting developments, given that the LLM will then try to re-contextualize the convo based on the info it has access to at that particular time... and again... and again.
Imagine someone forgetting the start of a conversation they are having with you, and then trying to continue and reframe your conversation without the earlier frame of reference, over and over again!
Case in point. My I, Caesar Battletech fanfic sits at 140,000 words by now. So I thought, hey, why not use ChatGPT to keep track of characters, events, and plots? I had it summarize the chapters, track the characters and events. Worked fine until halfway through the current size of the fic. Then it started to forget what came before, and started making shit up.
The kicker is: it never tells you as much. It never says: hey user, I've reached my capacity, whatever you add now will mean the earliest parts of our convo will be cut off.
Just let that sink in, and consider what this might mean for all those shmucks who use ChatGPT for some kind of scientific analysis or some such...
To answer my own question, having pumped a bunch of old prompts into MJ7 (as I do with each iteration of the system): no. If anything, the images seem worse than in v6.1, and maybe even worse than v5.2.
Video is a neat gimmick, but that's all it is. A better image editor with more and better features, for example for partial and/or gradual generation, for easier cleanup of noise, and an overall increase in detail through upscale would have been a way better way to spend their resources. Also, guns, gun turrets, missile launchers etc. are still a sore spot.
Video is sketchy, especially if you have more than one object moving, and you can see in real time how the models lose fidelity and quality as MJ frantically tries to extrapolate.
And yeah, the ladies in the videos? One is Cerys from my Marian Hegemony side-story, the other is Ava Kimura, from the main fic.
Video is a neat gimmick, but that's all it is. A better image editor with more and better features, for example for partial and/or gradual generation, for easier cleanup of noise, and an overall increase in detail through upscale would have been a way better way to spend their resources. Also, guns, gun turrets, missile launchers etc. are still a sore spot.
Video is sketchy, especially if you have more than one object moving, and you can see in real time how the models lose fidelity and quality as MJ frantically tries to extrapolate.
And yeah, the ladies in the videos? One is Cerys from my Marian Hegemony side-story, the other is Ava Kimura, from the main fic.
Experimenting with Suno:
I had ChatGPT o3 generate lyrics in Norwegian, including one in nonsense-verse with made-up words with Norwegian phonology. The song images were done with ChatGPT’s autoregressive image-gen.
Suno 4.5 is shockingly good. The old “Suno Echo” is gone and you can literally extend a song for 40+ minutes straight without it ever occurring.
I had ChatGPT o3 generate lyrics in Norwegian, including one in nonsense-verse with made-up words with Norwegian phonology. The song images were done with ChatGPT’s autoregressive image-gen.
Suno 4.5 is shockingly good. The old “Suno Echo” is gone and you can literally extend a song for 40+ minutes straight without it ever occurring.