|
Throwback Throwup by Unstable Confusion
[nfo]
|
||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
|||||||||||||
|
popularity : 58% |
|||||||||||||
alltime top: #8728 |
|
|||||||||||||
added on the 2022-10-23 01:17:58 by D.Fox  |
||||||||||||||
popularity helper
comments
well done! I feel thrown into an uncanny valley of 2002 demostyle and it's somehow flauschig <3
I like this, and so did my web browser.
Worthy winner of the compo, but I do feel it doesn't quite work out. You are getting great results out of the model, and it makes for a pretty coherent art style after heroic work picking them out, but I don't really think “making a demo about it” works when “it” is a specific technique for making images. (But it is very clearly marked both on the composlide and in the nfo, which it how it should be!)
It was said at some point—if you draw attention to an effect, it's no longer an effect. I've never before seen an effect with an overlay saying “wow, we spent so much time on this effect, it was really hard” :-) So I'd much prefer using AI art to make a demo to making a demo about AI art.
It was said at some point—if you draw attention to an effect, it's no longer an effect. I've never before seen an effect with an overlay saying “wow, we spent so much time on this effect, it was really hard” :-) So I'd much prefer using AI art to make a demo to making a demo about AI art.
Quote:
I've never before seen an effect with an overlay saying “wow, we spent so much time on this effect, it was really hard” :-)
You haven't watched a lot of 90s demos, have you? ;)
“Hey, look at this effect, we are really clever” is something entirely different.
I should probably follow up my snarky response with a serious one :) Using AI art to make a demo was actually our main intent, but we wanted to take the opportunity to also make a statement along the lines of "it's not as turnkey as some would like to believe" and that's why we also put the text about the amount of work in there.
And it's also a throwback to times when every effect had a panel along with it boasting the technology title ("interlaced quaternion fractal glenz vector morph with extra cheese!") and of course the amount of polys.
And it's also a throwback to times when every effect had a panel along with it boasting the technology title ("interlaced quaternion fractal glenz vector morph with extra cheese!") and of course the amount of polys.
Quote:
How many images did you need? (...)
It takes only 3-5 images of the object, and then the system is trained to "learn" the representation of the object in terms of its latent space, which as far as I understand is a long vector of abstract concepts and their contribution to the object in question.
So in a way the system learns that, say, D.Fox is 12.5% bear, 30% human and 10% cue ball.
This additional information is then merged into the representation of the prompt by the text encoder. None of the models making up SD itself are retrained or updated in the process.
The training process for Textual Inversion takes long because it needs to produce a sharp representation of the object with little extraneous noise. With too few training iterations we would get weird artifacts like every single image being set in a football stadium, I kid you not.
excellENT
For me it's very entertaining and stylish. Also it's an interesting attempt to explore the use of AI in future. AI is a new, somehow crazy way to do art but since it is very technical, it's totally demoscene related.
Fun and enjoyable. Not a fan of this whole AI art thing, but I thought it was presented in an interesting way. Cool music too.
Fun demo and fun theme. Also Im a sucker for the 95-02ish demostyle ;)
Nice idea, also very well executed!
nice
Happy cabling! Chill ch00n, funny comments and cool 👀-catchers which you obviously know that I enjoyed ❤️
congrats, nice one! thanks for publishing the patch!
ich rast komplett aus :D
And here's the cables patch for you to peruse. We even added a few comments!
Very fun usage of AI generated imagery! The demo itself was very entertaining and I enjoyed seeing this one live on the big screen too.
What NR4 said.
i loved the style, reminds me of my childhood :)
pouet comment in the style of Saga Musix, approving of this production, as a list of comma-separated sentences
Well done. Amazing work in mixing effects with gfx-design, good D&B+rasta music. Very Nice !
Entertaining!
I love you guys and gals. Thank you for letting me part of this fun prod! <3
"Diffusionist" 😁
Love the music. Not enough prods have banger DnB music nowadays :-)
lists containing this prod
submit changes
if this prod is a fake, some info is false or the download link is broken,
do not post about it in the comments, it will get lost.
instead, click here !
Interesting that you tried training the model with your own set of images. How many images did you need? Like 100 images each labeled with f.ex "D-Fox" or 1000 images? Do you add other tags like "sunglasses", "suit", "tie", "beard" etc?Were these images added to the other billions of images that is used to train the stable diffusion model? I suck at machine learning theory..