Views: 2074
Submissions: 24
Favs: 3
Digital Artist | Registered: June 8, 2016 06:10:35 AM
You may be wondering what in the world you are looking at. Let me explain.
Non Technical Introduction
To create the images you see, I take a very large number of images of some type (in this case, furry porn scraped from e621.net or derpibooru.org) and feed them to what's called a "machine learning" algorithm. This algorithm over time "learns" what makes these images look like they do, and attempts to replicate this to create new images. The current system I use I call ERIS because she relies mostly on random noise and doesn't actually have any understanding of what the images mean. Currently there are two versions of this system, ERIS-1 and ERIS-2. ERIS-1 produces 64x64px images, while ERIS-2 produces 128x128px images. In the future, I hope to develop more sophisticated algorithms that can produce higher quality and more specific images, perhaps even taking precise instructions (such as "A female cat anthro with extremely large breasts has vaginal sex with a male fox anthro with a large penis drawn cartoony") and producing appropriate images. See my journal on The Future for some ramblings on that. For some more technical details, see this journal.
WHY?
The good reason: e621 has an almost unprecedentedly well tagged and large gallery of images. Finding not only large amounts of data, but well tagged data, is one of the most challenging (if boring) parts of machine learning. In addition, I figured that the human eye is more likely to forgive flaws in drawn images than "real" images, so I hoped the results would be more aesthetically pleasing than trying to generate realistic imagery.
The real reason: I thought it'd be hilarious.
Are these images really created by an AI?
Sort of. The ones you currently see are created by a system I call ERIS, which uses a technique called machine learning. While similar to processes that happen in our human brains, ERIS is extremely simplified, and doesn't really have any "understanding" of images. On the other hand, ERIS was NOT programmed by a human, but rather "learned" by looking at thousands and thousands of images. Whether you want to call her an AI or not comes down to semantics.
Can you generate larger images?
No, ERIS-1 is built to generate 64x64 images and ERIS-2 is built to generate 128x128 images. I want to create other systems in the future capable of creating larger images, but doing so is prohibitive in terms of computation and memory requirements. I'm looking into clever tricks I could use, and will post my progress as it happens.
Can you generate a specific kind of image?
Not easily. ERIS has no concept of characters, angles, objects, themes or anything else. As such, all I do is feed her random noise and she produces random images. I can try to manipulate the output using some clever tricks, but this is a lot of work and I currently am not planning to look into that too much. In the future I will hopefully be able to make other systems that do understand context. See The Future.
Can I download this somewhere so I can make my own images?
Currently no, and this is unlikely to change. I do intend to release the ERIS-1 and ERIS-2 models (the "brains" so to speak) at some point, but actually using it is probably beyond the technical skills of most casual users. If you run 64 bit Linux, have a high end CUDA enabled NVIDIA GPU and intermediate programming skills, it would be possible to use this to get the same results I have. I have no intentions to create a "create_images.exe" type plug'n'play program at the moment, it would be a lot of work and making this kind of stuff run on Windows is hell. My time is better spent experimenting with better designs.
I really like an image you generated can I use it for XYZ?
Yes! I expressly grant permission to anyone to use ALL images in this gallery freely for any purpose, commercial or otherwise. Draw fanart of them, use them in your project, claim them as your own, sell them, I don't care one bit. I would love for you to send me an email at deepfur[at]hotmail.com if you do use any images, so I can see what you did with them, but this is not required.
I really appreciate your work and would love to help, is there anything I can do?
Said no one ever. Jokes aside, if you do want to help me create even more disturbing furry AIs, and have experience in machine learning, access to high end GPUs and/or disposable income, send me an email at deepfur[at]hotmail.com
I want to send you suggestions/questions/fan art/love letters/death threats, how can I contact you?
Send me a message here or send an email to deepfur[at]hotmail.com though I can't guarantee I can/want to answer every message.
Non Technical Introduction
To create the images you see, I take a very large number of images of some type (in this case, furry porn scraped from e621.net or derpibooru.org) and feed them to what's called a "machine learning" algorithm. This algorithm over time "learns" what makes these images look like they do, and attempts to replicate this to create new images. The current system I use I call ERIS because she relies mostly on random noise and doesn't actually have any understanding of what the images mean. Currently there are two versions of this system, ERIS-1 and ERIS-2. ERIS-1 produces 64x64px images, while ERIS-2 produces 128x128px images. In the future, I hope to develop more sophisticated algorithms that can produce higher quality and more specific images, perhaps even taking precise instructions (such as "A female cat anthro with extremely large breasts has vaginal sex with a male fox anthro with a large penis drawn cartoony") and producing appropriate images. See my journal on The Future for some ramblings on that. For some more technical details, see this journal.
WHY?
The good reason: e621 has an almost unprecedentedly well tagged and large gallery of images. Finding not only large amounts of data, but well tagged data, is one of the most challenging (if boring) parts of machine learning. In addition, I figured that the human eye is more likely to forgive flaws in drawn images than "real" images, so I hoped the results would be more aesthetically pleasing than trying to generate realistic imagery.
The real reason: I thought it'd be hilarious.
Are these images really created by an AI?
Sort of. The ones you currently see are created by a system I call ERIS, which uses a technique called machine learning. While similar to processes that happen in our human brains, ERIS is extremely simplified, and doesn't really have any "understanding" of images. On the other hand, ERIS was NOT programmed by a human, but rather "learned" by looking at thousands and thousands of images. Whether you want to call her an AI or not comes down to semantics.
Can you generate larger images?
No, ERIS-1 is built to generate 64x64 images and ERIS-2 is built to generate 128x128 images. I want to create other systems in the future capable of creating larger images, but doing so is prohibitive in terms of computation and memory requirements. I'm looking into clever tricks I could use, and will post my progress as it happens.
Can you generate a specific kind of image?
Not easily. ERIS has no concept of characters, angles, objects, themes or anything else. As such, all I do is feed her random noise and she produces random images. I can try to manipulate the output using some clever tricks, but this is a lot of work and I currently am not planning to look into that too much. In the future I will hopefully be able to make other systems that do understand context. See The Future.
Can I download this somewhere so I can make my own images?
Currently no, and this is unlikely to change. I do intend to release the ERIS-1 and ERIS-2 models (the "brains" so to speak) at some point, but actually using it is probably beyond the technical skills of most casual users. If you run 64 bit Linux, have a high end CUDA enabled NVIDIA GPU and intermediate programming skills, it would be possible to use this to get the same results I have. I have no intentions to create a "create_images.exe" type plug'n'play program at the moment, it would be a lot of work and making this kind of stuff run on Windows is hell. My time is better spent experimenting with better designs.
I really like an image you generated can I use it for XYZ?
Yes! I expressly grant permission to anyone to use ALL images in this gallery freely for any purpose, commercial or otherwise. Draw fanart of them, use them in your project, claim them as your own, sell them, I don't care one bit. I would love for you to send me an email at deepfur[at]hotmail.com if you do use any images, so I can see what you did with them, but this is not required.
I really appreciate your work and would love to help, is there anything I can do?
Said no one ever. Jokes aside, if you do want to help me create even more disturbing furry AIs, and have experience in machine learning, access to high end GPUs and/or disposable income, send me an email at deepfur[at]hotmail.com
I want to send you suggestions/questions/fan art/love letters/death threats, how can I contact you?
Send me a message here or send an email to deepfur[at]hotmail.com though I can't guarantee I can/want to answer every message.
Featured Submission
Favorites
This user has no favorites.
Recently Watched
Stats
Comments Earned: 18
Comments Made: 4
Journals: 5
Comments Made: 4
Journals: 5
Featured Journal
ERIS-2: MLP Edition: Take 2: Electric Boogaloo (G)
9 years ago
After the sort of mediocre results of the first attempt to train ERIS-2 on MLP images, I was itching to try a trick I had thought of. I separated the brightness value of each pixel from its color, and fed those values in separately. Otherwise, the two programs are identical.
As I had hoped, this had some very beneficial effects at first, making it easier for the AI to learn sharper edges and more distinct features (especially eyes!). But, bizarrely, it also caused it to occasionally create images that are basically just a single flat color with a little bit of noise, very strange. In addition, it seemed to create less and less interesting images as it continued to train, sadly. An interesting experiment nonetheless, I'll probably be doing this trick on all my AIs from now on. Though I think I have finally reached the very end of what the ERIS architecture is capable of, and so I have almost finished the next generation of AI architecture. It's a lot more sophisticated than ERIS, so I'm excited to see what it will do! As always, see this journal for the full technical rundown of ERIS.
Full training timeline: http://imgur.com/a/bq4au
As I had hoped, this had some very beneficial effects at first, making it easier for the AI to learn sharper edges and more distinct features (especially eyes!). But, bizarrely, it also caused it to occasionally create images that are basically just a single flat color with a little bit of noise, very strange. In addition, it seemed to create less and less interesting images as it continued to train, sadly. An interesting experiment nonetheless, I'll probably be doing this trick on all my AIs from now on. Though I think I have finally reached the very end of what the ERIS architecture is capable of, and so I have almost finished the next generation of AI architecture. It's a lot more sophisticated than ERIS, so I'm excited to see what it will do! As always, see this journal for the full technical rundown of ERIS.
Full training timeline: http://imgur.com/a/bq4au
User Profile
Accepting Trades
No Accepting Commissions
No
This user has not added any information to their profile.
PinkMew
~pinkmew
FA+