"A.I." (Machine Learning Algorithms) To Generate Art

Carnbot

There's loads of different types being developed. Scene photogrammetry from just a handful of photos
And 3d model generation from just one image

AudioGus

@Carnbot said:
There's loads of different types being developed. Scene photogrammetry from just a handful of photos
And 3d model generation from just one image

Yah lots of photogrammetry advances. I haven't seen anything along the lines of creative ideation yet though like you get with SD, MJ etc

Carnbot

@AudioGus said:

@Carnbot said:
There's loads of different types being developed. Scene photogrammetry from just a handful of photos
And 3d model generation from just one image

Yah lots of photogrammetry advances. I haven't seen anything along the lines of creative ideation yet though like you get with SD, MJ etc

I'm sure advanced algorithms are worked on for text to 3d and of course a skilled AI artist will be able to combine the two techniques eg create a photo then create the model from the photo.

AudioGus

@Carnbot said:

@AudioGus said:

@Carnbot said:
There's loads of different types being developed. Scene photogrammetry from just a handful of photos
And 3d model generation from just one image

Yah lots of photogrammetry advances. I haven't seen anything along the lines of creative ideation yet though like you get with SD, MJ etc

I'm sure advanced algorithms are worked on for text to 3d and of course a skilled AI artist will be able to combine the two techniques eg create a photo then create the model from the photo.

I do chop up 2D AI images in Maya. Am also looking forward to some gan styled morphy 3D sculpt stuff at some point, but I think it is a few years off for it to be worthwhile.

MadeofWax

If you want to get lost for a few hours seeing what has / can be created with Stable Diffusion and want to see the prompts that people use to create the images:
https://lexica.art/

auxmux

@MadeofWax said:
If you want to get lost for a few hours seeing what has / can be created with Stable Diffusion and want to see the prompts that people use to create the images:
https://lexica.art/

Nice, thanks, this will be helpful.

MadeofWax

Just so you can see that Stable Diffusion isn't just for art / illustrations:
Prompt: Long shot, Low shot, a group of 9 children trick or treating dressed as Slipknot in front of a house with Halloween Decorations, photograph

It's not perfect,(those Slipknot costumes could use some work) but you could fool people with the right prompt. Not to be a downer but people who fear these tools being in the wrong hands aren't overreacting. At the moment text to image converters seem to have issues with anatomy, especially hands. Sometimes you'll see extra fingers, extra or misplaced limbs and weirdly distorted faces. But given time and further refinement it will eventually be impossible to trust anything you see online.

MadeofWax

After seeing the cool comic stuff @echoopera did back on page 4
I wanted to give it a try.

Gilbert

That might be of interest.

Poppadocrock

Just stumbled across this on the App Store. Midjourney, expected release date October 1, 2022.

NeuM

@Poppadocrock said:
Just stumbled across this on the App Store. Midjourney, expected release date October 1, 2022.

Ah. Good one.

AudioGus

Apparently it is not by the same company. Probably won't make it to release from the sounds of it.

u0421793

In terms of music, this whole scene of transformers etc could have serious impact in terms of both composition and performance.

We think of Apple’s Drummer as AI, but is it really? I think the subject area indicated in this thread would give us some quite usable drummers, quite a relief to me (as I really am not interested in the drum side of my music, as long as it happens somehow).

AudioGus

NeuM

@u0421793 said:
In terms of music, this whole scene of transformers etc could have serious impact in terms of both composition and performance.

We think of Apple’s Drummer as AI, but is it really? I think the subject area indicated in this thread would give us some quite usable drummers, quite a relief to me (as I really am not interested in the drum side of my music, as long as it happens somehow).

Apple's Drummer is most definitely not "A.I." (and that is an overused term, I suppose because it's easier to say and remember instead of machine learning algorithm). Drummer is just cleverly designed and written software which allows a user to make a few adjustments between a limited number of settings and a grid so it fits a user's style of music. 🙂

At some point, we'll be able to invoke whichever backup software-based musicians we want to fill out a composition for us, or we'll give it a little direction and it'l figure out the rest.

waka_x

almost all my AI studies (using stable diffusion)-
various styles - thoughts and ideas...

http://tinyurl.com/25f6a7ut

follow my twitter if you like what I did- there will be more at some point...

Calverhall

I’m a bit obsessed with DiffusionBee, I lost the will to work on visual art decades ago, much prefer sweating over audio art. This does feel like a disruptive Pandora’s box moment however. Equal parts excitement and anxiety.

FastGhost

https://phantasmagoria.stavros.io/templates/

Poppadocrock

@AudioGus said:
Apparently it is not by the same company. Probably won't make it to release from the sounds of it.

Ah, really I thought it was.

Poppadocrock

I got an email that says Dall-E will now do faces.

jwmmakerofmusic

Dall-E is great, but uses a credits system per prompt. Mates, I downloaded an app called Wonder and purchased a lifetime license for 30 quid (otherwise it's $4.99 per week). (That money was to go to GSI apps, but I'll purchase those next month.) These are some of the images generated in Wonder...

😅

MrStochastic

@MadeofWax said:
If you want to get lost for a few hours seeing what has / can be created with Stable Diffusion and want to see the prompts that people use to create the images:
https://lexica.art/

It's interesting but it also seems like a window into our collective Id (Freudian) .

Listening to this podcast, https://www.radiolab.org/episodes/40000-recipes-murder , gets to just one unsettling aspect of AI. "Two scientists realize that the very same AI technology they have developed to discover medicines for rare diseases can also discover the most potent chemical weapons known to humankind. Inadvertently opening the Pandora’s Box of WMDs. What should they do now?"
Too late to put the genie back now.

AudioGus

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

NeuM

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

AudioGus

@NeuM said:

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

It doesn't really know how anything works. The whole thing is like a complex optical illusion in a way.

NeuM

@AudioGus said:

@NeuM said:

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

It doesn't really know how anything works. The whole thing is like a complex optical illusion in a way.

At some point it will solidify the connection that human beings generally have hands and fingers and that part of the model will be consistent. It's just a matter of "training" the dataset.

AudioGus

@NeuM said:

@AudioGus said:

@NeuM said:

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

It doesn't really know how anything works. The whole thing is like a complex optical illusion in a way.

At some point it will solidify the connection that human beings generally have hands and fingers and that part of the model will be consistent. It's just a matter of "training" the dataset.

A larger dataset will produce more convincing results but it really doesn't have a concept of humans or hands. I guess you are speaking more metaphorically though.

NeuM

@AudioGus said:

@NeuM said:

@AudioGus said:

@NeuM said:

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

It doesn't really know how anything works. The whole thing is like a complex optical illusion in a way.

At some point it will solidify the connection that human beings generally have hands and fingers and that part of the model will be consistent. It's just a matter of "training" the dataset.

A larger dataset will produce more convincing results but it really doesn't have a concept of humans or hands. I guess you are speaking more metaphorically though.

Right. It's just a matter of training the system. These systems are machine learning models, not genuine "artificial intelligence".

jwmmakerofmusic

All of these were AI generated with an app called Wonder.

None of these images of Samuel L Jackson are real photos but rather AI generated images! Set me back $30 for the perpetual license, but it helped cure my GAS for an entire month. 🤣

espiegel123

@NeuM said:

@AudioGus said:

@NeuM said:

@AudioGus said:

@NeuM said:

@AudioGus said:

They talk about Dreambooth which lets you train your own handful of images and get very impressive results. You can actually rent an online GPU for only a couple bucks an hour and essentially append your own images to a dataset for very little cost.

I wonder when these systems will be better trained to understand how hands and fingers work.

It doesn't really know how anything works. The whole thing is like a complex optical illusion in a way.

At some point it will solidify the connection that human beings generally have hands and fingers and that part of the model will be consistent. It's just a matter of "training" the dataset.

A larger dataset will produce more convincing results but it really doesn't have a concept of humans or hands. I guess you are speaking more metaphorically though.

Right. It's just a matter of training the system. These systems are machine learning models, not genuine "artificial intelligence".

AI is a term with a very broad meaning. Machine learning is a subset of AI. It is a particular strategy/technique within the more general field of AI .