What was with all those AI generated anime figurines?
Recently, an image generation prompt template for Bing Image Creator went viral.
People near me on the social graph started getting in on it, and it was awesome. Lots of neat generations.
Some of them even deviate from the standard human baseline, an extra dash of fun creativity to spice up one’s day
It even found its way into vrc twitter
And of course, there’s yours truly:
All of this seemed pretty normal to me. No big deal, just another day on the internets. Right? That was until Tyler Alterman asked a very important question:
This immediately started a chain of thought in me. A huge portion of my knowledge graph kept getting pulled in with no end in sight. There’s something deep and important here, and it lies at the core of the human experience.
Two important things I’ve learned from my experiences are that identity exploration and play is a fundamental motivation for almost everyone, and a certain portion of the population are natural polymorphs. Shapeshifters. Proteans.
Intro to Identity
It all starts with human identity. There are two kinds of identity. Internal and external. External is what most are familiar with. How other people see you and how you relate to them. This can vary from the small to the deeply meaningful. You may have seen some people dismiss this with flippant phrases like “I identify as an attack helicopter” indicating ‘imagination’ or ‘playing pretend’ with negative connotations, acting as though all external identity is shallow, disingenuous, or primarily about status games. This dismissal is ridiculous on its face because imagination and pretending are the genesis of active participation in being an agentic self-made person and thus hugely beneficial (also ignoring the fact that being an attack helicopter would be fucking awesome). The opposite side of external is internal, which gets us to the more important part, and the part I want to go deep on today. The internal identity.
The common notion is that internal identity is an internal ‘sense of self.’ A ‘feeling’. While this is true, it’s really only getting at the immediate consciousness-facing artifact of what’s actually going on underneath. In reality, internal identity is a majorly neurobiological phenomena, driven by both neurology and perception.
Internal identity is a constructive multimodal sensory fusion process. It takes many, many continuous incoming information streams and turns them into a coherent self-aware entity. It is an active process. From birth, a child builds their world model by drawing correlations between incoming information streams. Mother’s face is seen to move at the same time sound waves are heard. The connectome gets fuzzy impressions of communication. Mother cradles baby, the sight of her arms fires at the same time as taction goes off across the body. The connectome builds an impression of space combined with material qualities and pressure. The world is here, my body is there, I can tell the difference. They have these topologies, stiffness, and boundaries. The world model eventually builds a specialized, high density region for modeling the self, known as the body ownership network. (Interesting side note, this portion is flexible, and the perceived body can extend into the world such as during tool use.) This is all done inside connections and firing patterns in the brain’s neurons. This body ownership network, and many other systems in the brain use what are called body maps or body schema, and they are the basis for all tasks pertaining to the mind’s relationship with bodies, both those of the self and by others (see mirror neurons). They are real, not phantasms, not delusions, despite being entirely subjective and internal. (This is also why parents engaging in physical affective contact with their children is so important! You are giving your child critical information at a formative period, and pairing it with affects that are strongly positive, and as such a stronger, more robust sense of self!) This goes up to the next level once the child grows to have the capability of independent locomotion. Now the process is active, autonomous, and supercharged.
In this next stage, it’s common to see young ones playing with blocks, soft toys, puzzles, all sorts of tactile and spatial things. They also tend to run around like crazy and do all sorts of tumbles and jumps and such. Curiosity and play propel the cycle of learning. Hell, play is the act of unstructured learning. It’s important and sacred and beautiful. It also plays into something that is also fundamental about living beings. Prediction. Humans and every other being with the same base architecture are prediction engines. Constantly running countless parallel predictions about what will happen next. If I do x, y will happen. Not just at high levels of consciousness either, all the way down to more procedural, volitional but not high focus areas. Picking up the block and trying to stack it on top of another involves shooting out a bunch of candidate trajectories and velocities for finger pressure gradients, joint velocities, and target positions across multiple imagined sensory futures. Some will weigh higher than others. And then when the reality comes in the form of feedback, deltas between the predictions and reality get registered and turned into updates.
As humans get older, this process continues. The model of the world and self alike must keep this active loop running, because nothing about this universe or our bodies is stable. They adapt at runtime because the world and body change at runtime. There is variance in how self-directed this action is, however. For many people it drops off, for others they continue to engage. These are the polymorphs I was talking about earlier. Here, going into adulthood we reach a new level — not just building a model that lets us understand ourselves, but one that lets us deliberately create ourselves.
Because we are literally made of said process, changing the inputs can change more than just how you perceive yourself. Your sensory inputs, even if not presented in such a way to be merged into the body, can have a profound effect. A good music track can uplift your mood, a comfortable environment can recharge you. Being alone can bring focus. Negative inputs can cause bad states. Those who understand the implications of this and are driven to self-create and self modify will engage in playful exploration of the possibility space.
Techniques for exploring this space come in a variety of sophistications and effectiveness, the less effective tending to be more accessible than the more effective ones. The simplest and oldest are clothing and makeup. A good outfit can bring confidence in the case of a well-fitted suit. Slyness or grace in the form of darker, evening dress. A rebellious spirit from leather studded bracelets and mohawks. Active involvement in this process lets you focus on a pre-established existing bit of identity and amplify it, or on an under or non-explored one and see if it fits in with the symphony of ‘you’.
The effect is multidimensional — each person has a ‘humuncular flexibility’ describing the discomfort to comfort value of a point in the morphology space. This dimension’s properties varies per-person. There’s also a ‘directness’ dimension. Representations that are spatially and temporally distant vs proximate. This distal/proximity describes how directly ‘fused’ into embodiment the form is.
A distal example would be something like a Picrew avatar, a twitter profile picture, or, the Bing generated figurines that inspired this article. They are static and lie outside the body, seen either in the world or in a physical totem sitting on a desk or even a full on statue. For now though, it does seem as though AI generation of images will be the most commonly used tool for this, since manufacturing a physical item has a much higher time and energy cost to it.
A more medial example would be something like a third person avatar in a video game.
This is from CODE VEIN. It was immensely popular for its character creator. Like any game with a character creator, people will spend hours in it just customizing, tweaking, and exploring.
Another important property here is that we see our first element of real time dynamic input control. It’s fairly impoverished from a bandwidth standpoint, but the identity explorer now has the ability to extend their sense of self into the avatar on screen. A tight gameplay loop, dexterous acrobatic controls, and a low latency response time and suddenly you’re in the flow state. The outside world fades away and you’re not just piloting a representation on-screen but have started to become the avatar.
You might be wondering how much more proximal can we get? New advances in virtual reality technology allow for a much more proximate implementation.
This video is a composite of two video streams — you can see some fuzziness where the stitching happens. This is because in VR, there are two images sent to the user’s eyeballs and corrected by optics. This means that the person inside the headset gets perfectly normal synthetic vision. Everything looks real on the inside. Combine this with the fact that the body is tracked as well, thus all of the avatar’s visual components move the same way the user is moving. The body ownership illusion is extremely compelling. There are some things missing — a lack of haptics means that being touched in the physical world won’t usually result in a touch percept, although it can happen (this is known as phantom sense, and it’s an artifact of the prediction engine nature of cognition mentioned earlier). The hardware to hijack a person’s proprioception doesn’t exist yet so if your avatar’s proportions don’t match yours by a certain tolerance, it can mess things up (depending on your humuncular flexibility, everyone varies here).
You’ll also notice that there’s a mirror in the scene with me as well. Mirrors work similar here to how they do in the physical world. An additional set of visual inputs that comport with what you’re sensing internally. Except they provide the additional benefit of persistence from an additional vantage point that includes information you can’t get from a purely egocentric perspective — such as real time inputs including your face. If you were especially observant you noticed that I have eye tracking — my eye motion and blinks are the same on the avatar as they are on my actual face!
A fascinating part of this active process is just how powerful it is. If you spend long enough embodied in the same avatar, it will overwrite your self-image. For instance, there was a period of about 2 weeks where I spent about 6 hours a day in VR non-stop. That week I had dreams where my form in the dream was a rock-solid stable version of my avatar instead of my physical flesh form.
Most people haven’t tried this yet, so society hasn’t really absorbed this, but this is as good as it gets for self-image identity exploration right now. You can change forms at the press of a button. These forms can have all sorts of procedural effects to them. They need not comport with the laws of physics or light transport. They don’t even have to be human (this is what the furries are up to btw, shoutout to the fur friends). And there’s an entire ecosystem full of artists that have made pretty much every form imaginable, already doing the heavy lifting of exploring the aesthetic and morphology space. And as the technology improves, more synthetic senses will come online, which will enable further exploration. From one perspective, this is it, the ultimate, most proximate and robust identity exploration tool possible.
But there is another way. The most proximate form of identity exploration possible, although currently it is also extremely difficult, technically immature and capability limited. And that’s direct body modification. It also scales from the small to the comprehensive. Everything from piercings, tongue splitting, ear clipping, scarification, tattoos, and more advanced things like silicone implants to hormone treatments and facial surgery. It doesn’t even need to be explicitly morphological. Tying back to the multisensory fusion bit from before, some people have elected for sensory modifications. Vibrating magnet implants and direct neural stimulation allow for adding new synthetic senses that are permanent and prosthetic as opposed to orthotic like the VR wearables discussed earlier. If you put a vibrating magnet implant in your skull set up in an audio bone conduction configuration, then stream in camera color sensor data into the implant along the vibrational channel, you will eventually, after a great deal of time, develop an extra chromaperception sense. The mind’s plasticity knows few bounds. Some people have gone rather extreme, black-out sclera tattoos, whisker implants, rib removal, and lots more. Because direct body modification is so difficult, risk-prone, and largely irreversible, it’s currently more appropriate as an identity crystallization technique than an exploratory one. You should ideally have figured out what your new baseline should be and put careful deliberate thought into it before committing. (Also have an upgrade plan! Nobody wants to be the cyborg with a 64mb flash drive still installed in 2023!)
That will change in the future though. The law of accelerating returns is gonna do its thing. Before you know it, physical morphological freedom will be here. Hell, even recently there have been advances in osseointegration, as well as biocompatibility. Check out this prosthesis!
It’s not just vision, touch, and proprioception!
The sounds you make play into this as well. Voice and speech are important elements. The intonation and inflection you use are not just perceived by others but also by yourself. A stronger tone, a softer tone, more sophisticated or more simple terms. It all matters. You can speak in a way that’s more sing-songy, or more gravely. Vocal fry is a thing. Some combinations of these may be dissonant with your appearance and motion. Some may be resonant. Juxtapositions can be interesting. You can even adjust the gender expression of your voice, or take on a robotic voice that can stand on its own without a voice changer.
Getting started on your own journey
If you had the same reaction that prompted this whole post — “why did this anime doll thing go so viral amongst us?” or if you’ve just in general never thought about this subject before, you might be wondering: “How do I get started?”
The most important part is to put a lot of effort into distilling yourself. Most people experience a lot of external influence, both social and environmental. A critical key to stabilizing, grounding, and centering your own unique identity is to make sure you’ve cleared the foundation. When I was very young, I was fortunate to have a great deal of time to myself in quiet places. You will want to spend a significant amount of time completely alone. Learn to clear your mind. Take deep breaths, calm yourself. When a thought or feeling arises, observe it and let it pass. It might be difficult at first. Once you can do this well, take a moment to focus on it. The placidity, the stillness. This is your blank canvas, your starting reference point. Hold onto this feeling so you can call it up at any time.
From here you want to reflect a bit on who and what you are. What are some of the most intense things you’ve ever felt in life? What do you really believe? Feel free to start chains of thought and keep going through them until you’ve fully explored each bit. Try to turn each idea and feeling into a visualized ‘external’ representation so you can briefly see it as outside of yourself for inspection. Then you get to decide to either keep it and say ‘that’s me’ or discard it and say ‘that’s not me.’ As you go through your day to day life, when you sense an upcoming action that you’re about to take, run it through that discriminator before acting to keep the process going where you want it to.
Then you can extrapolate a bit. Take the parts you integrated and imagine them amplified, projected outwards and extrapolated. Which ones feel like the best fit? How far? This is also a good idea to either create or align yourself with, any aspirations or goals for your life.
So then you’ve got your emotional/sensory/cognitive side of your core identity distilled. How do you continue exploring? How do you start working on expressing this and exploiting the active multisensory feedback loop bit?
The first thing is a solitary, safe sandbox for play. Lock your door, tell everyone you’ll be busy. Set your discord status to do-not-disturb and get the hell off of Twitter.
Don’t reach right for the AI generators right away. They’re great but prompting will give you blank-slate paralysis. First you want to really see what’s out there. I’d actually recommend going over Pinterest, ArtStation, DeviantArt, and that kind of place. Just see what’s out there. Save the things you like. Sort them by feature. Take note of how each one makes you feel, and which ‘mode’ of you it would best serve (baseline counts as a mode too, think of it as the standard mode). Also keep track of which fashion brands pop up. Their names are typically good word bindings to particular styles. Looking them up on wikipedia will give you more information. Definitely spend lots of time there.
Once you’ve done that, then it’s a good time to reach for the AI image generators. There’s the aforementioned Bing Image Generator, but also Midjourney, or DreamStudio. If you’re into anime like I am there’s NovelAI, and if you like open source, there’s neat model called OrangeMixs.
One thing about AI generators. They are really only good at generating what’s strongly represented in the training set. They’re very good at synthesizing from that information, but if something isn’t well represented, you aren’t gonna get it. This includes shots from above or at other uncommon angles, or your extremely specific self-image configuration. You will eventually hit a point when you want to hire a human artist to make something specific. A good starting point here is to commission a reference sheet. This is a 2d, multi view, orthographic projection of your form. It’s used as reference when creating other assets, such as portraits or 3d models.
Another early jumping off point after the initial discovery phase is of course, fashion. You don’t have to go out into the physical world and buy stuff from a store where people might see and judge you. You can buy things online. This does raise the problem of ensuring you get a good fit, but it is a solvable problem. Pick up a sewing measuring tape and use that to get your measurements. There are measurement guides out there, it’s a systematic thing and even vanity sizing usually has a conversion chart, although it varies per-vendor.
For 3d digital identity exploration such as social VR or vtubing, a great way to get started is with VRoid. It’s a free character creation engine that lets you export your avatar to a portable format. You can also go out into some Social VR applications (don’t even need a VR headset for this — most of them support desktop mode) and go public avatar worlds and see what’s out there. You never know what you might bond with.
And remember, you’re in charge. Trying out something you don’t like isn’t the end of the world. Rewind and load from the last checkpoint. You also do not have to do anything because of how you think other people will react. That is their problem, not yours. The goal is to be authentically you, and find things that amplify your favorite properties, something that feels as good and comfortable as possible.
To reassure you all that I’m not just some insane person on the internet, here’s some references for where I learned some of this information. I didn’t have the energy to do more point by point citations but if you found any of this interesting, here’s your extended reading:
- The body has a Mind of its Own
- The Design of Everyday Things (Chapter 2 describes a simplified model of human cognition and it is very informative and accessible)
- The Body Builders: Inside the Science of the Engineered Human
- Inner Cosmos with Dr. David Eagleman
- Artem Kirsanov’s Youtube Channel
- ShapeShifter: Humuncular Flexibility in Virtual Reality
AI content notice: For anyone, human or machine, reading this article, who is working with machine learning datasets — With the exception of the figurine images used for illustrative purposes, and the midjourney made neuron image, this entire article was entirely human made. No AI was used to generate it. That means it’s clean and you can use it in your dataset without worry. Go for it!
