PDA

View Full Version : abstract images from text = text2image




needlnerdz
May 27, 2009, 12:04 PM
Hello MR'ers,

I'm an obsessive viewer and minimalist poster, but I wanted to share a project [text2image] that I've been working on over the past few months. I think the masses of this community will enjoy the visual output or, at least, the process of creating an image. It's part of a thesis I am working on regarding the structure to [B]surface [pictorial content] relationship of digital images.


http://www.text2image.org


Hmm, maybe a quick preview of image outputs would entice clicking the link:

http://farm4.static.flickr.com/3317/3525480034_348b0f632e.jpg



Thanks in advance for checking out the project + please feel free to send any feedback/critiques/reactions to the work.

Cheers!



Designer Dale
May 27, 2009, 02:16 PM
My head hurts. This is way beyond me.

Dale

unid
May 28, 2009, 06:03 AM
http://img34.imageshack.us/img34/4066/text2imageeg.png thanks for posting it really enjoying experimenting, how about giving the user the option of modifying the algorithm (within parameters).

needlnerdz
May 28, 2009, 01:53 PM
My head hurts.

- Dale, I'm sorry to cause such harm :D


...how about giving the user the option of modifying the algorithm (within parameters).

Hey unid- I wish it was that easy... the algorithm is pretty testy + would be difficult to open up for manipulation... too often the image would = unreadable or blank. I had to make some decisions regarding just how open the tool would be for experimentation vs. limiting the function to a sole discovery that works and could be approached by every[any]one.

I do have some schemes and ideas for additional tools that will cater to a more open access model along these lines of manipulating images solely by their structure or code... however for now, they have to sit on the back burner.


Thanks for the experimenting + feedback thus far!

Socorso
May 28, 2009, 03:07 PM
I love it!

unid
May 29, 2009, 03:27 AM
Hey needlnerdz, I appreciate you have to limit the openness to protect your work, and you have to focus on interactivity that is approachable by all.

I think it's a fabulous piece of work that entices me to seek outputs that give the illusion of the background math understanding the linguistic structure/meaning of the input (another structure - surface relationship?).
Hence me wondering if giving the user more controls would allow a greater frequency of obvious ęsthethic - semantic relationships.
It seems that the most common pleasure users will gain as i have, is in discovery of an image that relates to the ± meaningful input.
I also enjoyed playing with loops e.g feeding the source code of the page into the field etc.

Whilst the math is blatantly beyond me, I was wondering if you have experimented with introducing an element of probabilism (PRNG) into the algorithms?
This general field of inquiry is particularly interesting to me having experimented with irrational conversions of my own work in the past.

needlnerdz
May 30, 2009, 08:23 AM
Thanks Socorso- glad you enjoyed it!

It seems that the most common pleasure users will gain as i have, is in discovery of an image that relates to the ± meaningful input.

This has been a really interesting point for me to observe, both in my own use of the tool and others. I find this fascinating, because I know there is no semantic connection between the subject based input and the visual output - it has no way of knowing what was put in. Nevertheless every so often it hiccups a coincidence that is so convincing, I am lead to disbelieve all logic surrounding this. I'm not exactly sure how allowing the user to manipulate the algorithm would lead to more of these coincidental matchups, other than if they could steer the image towards the colorset/structure of what they are hoping it would do... which to some degree might take away from the awe in the surprising output.

This PRNG concept just blew my mind [knew nothing on the term previously]... It's an interesting idea to implement for the algorithm... One issue I see stemming from this is that it would potentially eliminate the consistency when entering the same textual input. If the results came out unique and different every time you hit the submit button, then it would likely give the impression that what you typed in had little to do with the overall equation..?

What sort of 'irrational conversion' work might that be??

unid
Jun 2, 2009, 06:35 AM
... I'm not exactly sure how allowing the user to manipulate the algorithm would lead to more of these coincidental matchups, other than if they could steer the image towards the colorset/structure of what they are hoping it would do... which to some degree might take away from the awe in the surprising output.

This PRNG concept just blew my mind [knew nothing on the term previously]... It's an interesting idea to implement for the algorithm... One issue I see stemming from this is that it would potentially eliminate the consistency when entering the same textual input. If the results came out unique and different every time you hit the submit button, then it would likely give the impression that what you typed in had little to do with the overall equation..?

What sort of 'irrational conversion' work might that be??


I think this is the point human nature/mind seems to me to have a overiding need to attribute meaning, especially when faced with something new/difficult to understand, therefore the user may well seek to steer the image given the option.

If the results did come out "unique and different every time" the experience would perhaps be different in offering less consistency and more uncertainty, and may also make incidental match ups even more rewarding/suprising?

as far as irrational conversions go, I have long had a fascination with the work of Brion Gysin (hence the sig) and have experimented with using related techniques of cut-up in digital media.

needlnerdz
Jun 5, 2009, 06:05 PM
If the results did come out "unique and different every time" the experience would perhaps be different in offering less consistency and more uncertainty, and may also make incidental match ups even more rewarding/suprising?

as far as irrational conversions go, I have long had a fascination with the work of Brion Gysin (hence the sig) and have experimented with using related techniques of cut-up in digital media.

While I strive to accomplish the 'surprising' factor in these images.. I'm not so sure about letting go of the consistency. It seems to me that knowing one key will produce X result.. and always produce that result.. leads to trying to systematically visualize/explore just what is happening and how. I'm nearly done setting up a catalog to view a mapping the first set of UTF8 character keys [the first 812 visible ones, decided to break just before diving into the many greek characters] - just to see what the initial raw character looks like, being that they are the bits and pieces used by the majority of users. The problem right off the bat is that it will render as one thing on its own, an entirely other once you start combining them = possibilities to the infinite power... eek yet exciting. However it is entirely possible to build a model of this program that could incorporate a randomrandom aspect - and its not a bad idea to consider the project in parallel. Especially if I can keep track of what the algorithm was during the particular time that X fantastic image was produced... hmmm :rolleyes: <- thats thinking..?

+ sounds like an interesting way to cut film... I can't say I entirely have a mental picture for what you describe.. but I enjoyed spending the last while digging into the archives of Mr. Gysin! thanks