|
Dithputation by Team210 [web]
[nfo]
|
||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
|||||||||||||
|
popularity : 53% |
|||||||||||||
alltime top: #52064 |
|
|||||||||||||
|
||||||||||||||
added on the 2024-08-18 13:22:59 by NR4  |
||||||||||||||
popularity helper
comments
AI pixels.
sucks added on the 2024-08-18 13:42:52 by Preacher 
nice
Quote:
AI pixels.
You kinda missed the point of the prod tho. I was showing off the pixel art converter I wrote.
I am sure that in the entirety of the demoscene there would've been an artist that would have provided you some actual art. You even acknowledge that by writing "stolen from many artists at once. Sorry for that." in the infofile.
Yeah, no. The tech stack may be nice from what I can tell, but it's still poison.
Yeah, no. The tech stack may be nice from what I can tell, but it's still poison.
I agree with both of your points tbh, that's why I disclosed it on the beamslide and in the nfo. And that's why I didn't try to enter something in the pixel art compo, which would have easily been possible using the tool.
I mean, 4k exe gfx _is_ a competition for coders. I believe it's suitable there. I will try and find real artists to use it however, if I use it again.
I mean, 4k exe gfx _is_ a competition for coders. I believe it's suitable there. I will try and find real artists to use it however, if I use it again.
my "issue" with this entry was that the image looks as if it would fit just fine in 4k regardless of how it was generated or converted
AI thumb
Ah what the hell, this prod was a bad idea, now that I think of it. The whole workflow is inherently shit, and the idea of showing off an algorithm in a compo where most people want to compare art, not algorithms, seems idiotic to me now. Sorry for wasting your time.
Thought about releasing the tool as well, but that one will stay private.
Thought about releasing the tool as well, but that one will stay private.
if you're extrapolating from people's comments that the tool isn't worth releasing i'm pretty sure you're misinterpreting people's intentions
What havoc said.
This thumb is for AI usage only, not for the underlying tech, which i’d love to see used for content made by actual human beings.
This thumb is for AI usage only, not for the underlying tech, which i’d love to see used for content made by actual human beings.
What Uncle-x said
i think the idea is awesome and i second Preacher here, but i won't thumb it down for technical reasons. Maybe make another version?
Thumbs up just to make up for NR4's thumbs down (a thumbs down done for the wrong reasons since ART is not something you do to please others but, if anything, to make them think and provoke them).
And code is ART too.
And code is ART too.
I like the algo but this particular type of AI usage in demoscene makes me wanna vomit.
sorry for this, too
Quote:
but this particular type of AI usage in demoscene makes me wanna vomit.
what exactly are you implying?
@NR4: I think when people use expressions like "this makes me vomit" it's because there is a visceral reaction. That is, purely emotional and without rational mediation. Therefore, you will not get anything by trying to convince them. The best thing you can do, in my opinion, is to publish the tool and the paper and let them rage.
Art is, at the moment, purely human. Generative AI is just a tool that makes guesses.
In a few years, all graphic artists will use AI tools as part of their workflow. All this general anger against AI will be seen as nonsense.
Art is, at the moment, purely human. Generative AI is just a tool that makes guesses.
In a few years, all graphic artists will use AI tools as part of their workflow. All this general anger against AI will be seen as nonsense.
Awesome job! can't wait to read the details about the tool. Huge thumbs up, please ignore the Anti-AI Mob that doesn't vote based on reason but rather from visceral reactions and fear of the unknown.
A good dithering algo is always welcomed. I hope NR4 ignores the haters and publish some info about it :)
not giving a thumbs down because i know that this entry was primarily about your tool dithertation, which, from all i've seen, is really cool and i'd love to play around with it sometime! but while generic GenAI anime girls might be ok for testing, their inclusion in an entry kind of ruined what would've otherwise been a fun showcase of dithering algos and image compression in 4k for me. collaboration with a scene graphician and maybe a more interesting subject matter could've easily made this one of my compo favs.
what peskanov said... but also what Preacher said... but also what havoc said!!
counterthumb
I'm really a bit entertained by the amount of comedy in this prod thread.
* I spent more time on this prod than I ever have on any other executable gfx. Constructing the algorithm, writing the tool and learning to use it took 3 Months, 3-4 hours a day.
* The prod has visibly a lot more to it than simply the GenAI input (How good is GenAI at pixeling correctly, really? Also, result is packed into a linux executable).
* I like how there is a woman in the input picture and instantly some try to make this thread about politics. Nobody cared when I showed a body builder.
My position on this:
* Sorry for falsely assuming that I am welcome to enter a scene competition with AI if my process is sufficiently advanced.
* The subject of the input picture is very far from important for the concept of this prod, yet definitely not up for discussion. If you find it necessary to make a clear technical demonstration about politics, then so be it. Puke all day if it makes you feel better.
* I'm glad there are actually people who are interested in the algorithm & I will release it eventually.
* I spent more time on this prod than I ever have on any other executable gfx. Constructing the algorithm, writing the tool and learning to use it took 3 Months, 3-4 hours a day.
* The prod has visibly a lot more to it than simply the GenAI input (How good is GenAI at pixeling correctly, really? Also, result is packed into a linux executable).
* I like how there is a woman in the input picture and instantly some try to make this thread about politics. Nobody cared when I showed a body builder.
My position on this:
* Sorry for falsely assuming that I am welcome to enter a scene competition with AI if my process is sufficiently advanced.
* The subject of the input picture is very far from important for the concept of this prod, yet definitely not up for discussion. If you find it necessary to make a clear technical demonstration about politics, then so be it. Puke all day if it makes you feel better.
* I'm glad there are actually people who are interested in the algorithm & I will release it eventually.
@NR4, you're mentioning you did a packer for pixel art, does it mean that your code it not really about generating this specific picture, but is more a generic pixel art packer? That would be very interesting (well somewhere between very interesting and fuckin' awesome). Could you post other pictures than this one packed using your system? The pic you used looks very RLE-packable, is that a constraint?
I think that you shouldn't have been so surprised about getting a negative reaction to using AI (and the subject matter) in the demoscene by now.
Anyway, some thoughts and questions:
Anyway, some thoughts and questions:
- At first I thought this was on an oldschool platform. Is the whole decoding process contained in the executable, and could it be rewritten for Amiga etc.?
- From just looking at the image with my eyes it seems like it has a lot of regions of solid background colour, which I imagine compress pretty well. The picture itself hides this pretty well. Did you pre-process the image "manually"?
- Regarding the use of AI, it was pointed out by Preacher that you could have asked an artist to draw it for you - this wouldn't have been feasible if you were somehow searching through a large space of generated images to optimise for compressibility. Were you doing something similar?
soundy: yes exactly - it takes any input picture and dithers it in a way that can be used to achieve a pixel-art impression (which is a slightly different task than simply dither an image as accurately as possible; a task which the tool performs quite ok tho as well).The specific mask I used is optimized to pack well under epoqe's binary encoder; I assume that adapting to different packer mechanisms is possible with not too much effort tho.
fizzer: the decoding process is x86/linux specific and would not be possible on amiga imo, maybe it would be possible to find a mechanism viable for amiga sizecoding tho. The image is converted automatically without my manual intervention, but part of the algo is a certain preprocessing (changing saturation, gamma, contrast, ...). The conversion is possible for any input image, but some styles produce results that look more like pixel art than others; for example fotos do usually not convert into convincing pixel art while cartoon-like drawings or paintings work better. It would be something that the artist would have to try, what techniques work best :) hand-drawn styles seem to work in general
fizzer: the decoding process is x86/linux specific and would not be possible on amiga imo, maybe it would be possible to find a mechanism viable for amiga sizecoding tho. The image is converted automatically without my manual intervention, but part of the algo is a certain preprocessing (changing saturation, gamma, contrast, ...). The conversion is possible for any input image, but some styles produce results that look more like pixel art than others; for example fotos do usually not convert into convincing pixel art while cartoon-like drawings or paintings work better. It would be something that the artist would have to try, what techniques work best :) hand-drawn styles seem to work in general
I'm happy to test the tool for packing content into oldschool prods if any of you are interested! :) And maybe also modify it to meet specific hardware requirements, like for example it can already convert to the C64 palette, but obviously a C64 could not render the results because of the additional limitations what it can display.
Imho the issue with this prod is really just that it's in the wrong compo (due to the lack of a fitting one) you already said it:
Like havoc pointed out presenting the image doesn't appear like a technical challenge either. Considering its 4 color palette, native resolution and scanline style dithering, using a compacted (2+6bit) rle the raw image data comes down to 3862 bytes including the palette(true color) and resolution (...yes I couldn't help myself).
So this prod isn't about the image, nor is it about the encoding of the image, so please do the writeup about the algo so we can see what this prod is actually about. ;)
Motivational and entertainment thumb
Quote:
showing off an algorithm in a compo where most people want to compare art, not algorithms
Like havoc pointed out presenting the image doesn't appear like a technical challenge either. Considering its 4 color palette, native resolution and scanline style dithering, using a compacted (2+6bit) rle the raw image data comes down to 3862 bytes including the palette(true color) and resolution (...yes I couldn't help myself).
So this prod isn't about the image, nor is it about the encoding of the image, so please do the writeup about the algo so we can see what this prod is actually about. ;)
Motivational and entertainment thumb
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 !