AI ART GENERATION
Giving birth to art that never existed
The Art
Replicant Army is an AI-art generation studio with an original collection of 10,000 fictitious humans and other works.
The Mission
Question the limitation of reality and how art and society can interact with machine learning technology (AI) to create new forms of art and media.
The Opportunity
Explore the topic of digital identity and synthetic people manufacturing with art, along with the creation of other collections that explore the limitations of AI generation.
The Why
No one can ever 'uninvent' a technology, so we need to either embrace it or get out of the way, we choose to embrace.
The Master Source
Latest Creations Here
HTTPs://www.Pond5.com/Artist/ReplicantArmy (Enjoy 20% off on all media from any collection!)
The Newest Collections
The Original 10,000 Fictitious Humans
We used the latest algorithms to create 10,000 unique Replicants that looks like humans, but were not taken from photographs of actual humans. These are 100% synthetically created humans where most look identical to real people, that said there are some glitches and rare attributes to keep it fun. Note the distribution with the percentages below each category.
The probability of 2 replicants having identical attributes is: 1 in 241,674,000,000,000 ...that's 1 in every 241 Trillion created!!!
10,000 Fictitious Humans
A collection of fictitious humans created as illustrations and 3d models.
3D Models
Gender
All Replicants are a binary classification of Female or Male but you'll find the lines are very blurred as to what the AI generation thinks.
(See confidence score below for more details)
Female
56.49%
Male
43.51%
Age
Replicants span the full age range of approximately 1 to 80 years with a binary classification of Child or Adult.
Child
11.72%
Adult
88.28%
Race
Replicants are split into 3 races, where a best guess is made based on skin tone and other facial markers and each has a confidence score.
Negroid
7.36%
Mongoloid
12.56%
Caucasoid
80.08%
Hair Colors
224 Different Hair Color Groupings from 8002 Unique Colors
Most popular color:
Raisin Black 15.8%
Skin Tones
136 Different Skin Tone Groupings from 9170 Unique Colors
Most popular color:
Light Taupe 6.1%
Eye Hues
95 Eye Hue Groupings from 8317 Unique Colors
Most popular color:
Dark Sienna 2.4%
Hairstyles
There are 33 unique types of hairstyles with 6 levels (Common, Rare, Epic, Legend, Artifact, Heirloom) and some are very rare.
Common Hairstyles = 5789/10000
High Volume Brushed Up
28.18%
Long Hair 2
10.02%
Long Wavy
5.44%
Mid Length Ruffled
5.23%
Mid Length Straight
4.72%
Long Crimped
4.30%
Rare Hairstyles = 2251/10000
Long Bob
4.17%
Short Curls
4.00%
Mid Length Wispy
3.78%
Very Long
3.65%
Classic Side Part
3.48%
Long Hair
3.43%
Epic Hairstyles = 1488/10000
Crew Cut
3.35%
Wavy Shag
2.95%
Short Disheveled
2.95%
Straight Bob Bangs
2.60%
Balding
1.78%
Curtained Hair
1.25%
Legend Hairstyles = 330/10000
Ponytail with Bangs
0.93%
Classic Taper
0.53%
Absolutely Bald
0.53%
Short Bob Asymmetrical Bangs
0.49%
Side French Braid
0.42%
Crew Cut 2
0.40%
Artifact Hairstyles = 132/10000
Short Simple
0.28%
Crew Cut 3
0.27%
Long Afro
0.26%
Cowlick
0.18%
Flat Top
0.17%
Cowlick 2
0.16%
Heirloom Hairstyles = 10/10000
Long Disheveled
0.04%
Bob
0.04%
Burr Cut
0.02%
"They Look So Real!"
Each replicant was created using a generative adversary network (GAN). This technology uses machine learning algorithms to make massive calculations to generate life-like people. These are not actual human people that have physically lived on this earth. All were created to simulate a real human being, and as real as they look, they are not actual based on images of real people. If any picture looks like you or someone you know it is 100% coincidence, as there is a finite amount of combinations to make a human. Finite being a very large number...but still finite.
How a Replicant is Created
The technology used to create AI generated virtual humanoids is called a generative adversary network (GAN). Here is a quick overview of how it works from Norman Ponte (https://www.zumolabs.ai/post/what-are-gans) to get you up to speed.
Otherwise the TLDR version is: "TONS of photos were used to 'train or teach' the computer what a human looks like and then the computer uses its best guess to create a new original photo".
Generative Adversarial Networks, or GANs, are an implementation of a Generative Model [Figure 1]. A generative model is a neural network which is trained to output a new example (e.g. an image) from a distribution of provided examples.
GANs are composed of two parts: the Generator and the Discriminator [Figure 2]. The Generator and the Discriminator are two separate neural networks: their goal is to transform a given input into a desired output.
The Generator is a generative model: given a random vector of noise this network will generate something which is within the distribution of the training data. The Discriminator is a model trained to return the probability that the input example came from the original training data—that is, is from the same statistical distribution—and was not just randomly generated [Figure 3]. Think of the Generator as an art forger and the discriminator as an art evaluator.
Both of these models are trained to achieve that same end goal by using back propagation. Back propagation is the process of iteratively moving the weights of the neural network towards the desired goal [Figure 4]. The Generator and the Discriminator (borrowing some concepts from game theory) are placed as adversaries, allowing them to train each other to be better at their respective tasks: generating and discriminating. Though trained in tandem, the ultimate goal is really to train a strong generative model.
So as you can see the Discriminator is a key piece of this that determines if the final image looks like a human or not. If this doesn't wet your beak enough and you really want to geek-out on this technology, checkout this paper as it explains the algorithms in more detail: https://arxiv.org/pdf/1406.2661.pdf
The Support
We are mere torch carriers passing the last generation's innovations to the next, as no man is an island, and the technologies of the past need to be celebrated as the influencers they are. Here are the people and projects that made this possible that need to be recognized and praised as mostly direct contributions.
Len Sassaman
Huge contributor to cryptography, open source and possibly is Satoshi Nakamoto (creator of Bitcoin)...more here.
Dan Kaminsky
Epic internet security expert and probably the best conference speaker for geeks around.
StyleGan2
Ground breaking project in creating lifelike machine learning people https://github.com/NVlabs/stylegan2
AvatarSDK
Incredible 3D modeling software for mapping human faces. https://avatarsdk.com/
ARweave
The permaweb company that allows us to store these Replicants indefinitely. https://www.arweave.org/
ARdrive-cli
Project built on ARweave that allows for clean management of data assets. https://github.com/ardriveapp/ardrive-cli
ARdrive-get-files
Simple script to get all original filenames and ARweave data tx IDs in a CSV. https://github.com/Silanael/ardrive-get-files
Python 3.10
Core programming language for data manipulation and other cool stuff. https://www.python.org/
Node.js
Great collection of software tools for solving problems. https://nodejs.org/
PNG Gauntlet
Nice simple tool for optimizing PNG files. https://pnggauntlet.com/
ImportJSON
Sweet script for importing JSON files into spreadsheets https://github.com/bradjasper/ImportJSON
Bulk Rename Utility
Awesome tool for renaming files in bulk with regular expressions https://www.bulkrenameutility.co.uk/
CopyFilenames
Quick extension to bulk copy filenames https://www.extrabit.com/copyfilenames
FBX2GLFT
Tool for converting FBX files into GLB. https://github.com/facebookincubator/FBX2glTF/
Autodesk 3D Max
Gold standard for manipulating 3D models and animations https://www.autodesk.com/products/3ds-max/overview
QGifer
Simple animated GIF maker from video files https://sourceforge.net/projects/qgifer/
Macro Recorder
Powerful macro maker for when you can't code it yourself https://www.macrorecorder.com/
STL-Thumb
Thumbnail maker for STL files https://github.com/unlimitedbacon/stl-thumb
Meshtool
For manipulating meshes in a 3D model https://github.com/pycollada/meshtool
Stable Diffusion
For creating next generation AI images https://github.com/CompVis/stable-diffusion
Topaz Gigapixel AI
For upscaling images https://www.topazlabs.com/gigapixel-ai
The Fun
Get your Replicant physically 3D printed with the latest full color technology, it's awesome!
The Artist
Phil Maher
Computer Scientist, Software Developer, Internet, Open Source, Entrepreneur, Bitcoin, Drone Cinematographer, Stock Footage, Film Archivist, Writer, Builder, Artist.