How does creative design work?

[holland]

When we talk about skills like character design, fashion desgn, prop design, environment design... we are fundamentally talking about multiple choice problems. Each design we make is merely a series of prompts with a set of potential options to choose between.

For example, when designing a character's face, we might choose between these options when deciding the distance between their eyes:

• their eyes are close together
• their eyes are an average distance apart
• their eyes are far apart

Each artist will have their own sense of what each option looks like. Some may have more or different options in mind -- like adding "their eyes are slightly far apart" and "their eyes are slightly close together" for additional nuance. Others may have a single preset eye distance with no variance between characters.

Neither the questions nor the solutions are set-in-stone objective matters. Everyone's "multiple choice test" in this metaphor is uniquely their own.

Even when an artist isn't consciously trying to think with a framework like this when designing, the brain is still going through that process. So consciously acknowledging it and laying bare what those questions and answers are provides a variety of benefits:

• We get a clearer sense of what we're doing. The design process feels more intentional and specific, as opposed to a mushy "trial-and-error" experience. Besides trial-and-error often leading to more errors than successes in beginners, it can also lead to us falling back on repetitive defaults or overdone tropes without even realizing we're doing it.
  
  
• We can clearly see the gaps in our knowledge and its application. For example, you might feel that you have a very intimate understanding of eye anatomy at some conceptual level, yet you discover you have no idea how far you can push the eye placement in any direction nor how you can actually vary its features. So what exactly was all that anatomy for?
  
  
• We have a place to cement our studies into our process. While we may quickly outgrow the need to check explicit option lists just to whip together a design, keeping them around lets us log the new things that we learn about in that area of design. It can also be used to refresh ourselves periodically, where it's common to find a cool option we completely forgot about and have been neglecting in recent work.
  
  
• We can more easily dissect existing designs and other styles that interest us. It's more than a little frustrating to be looking at an artist's work and going, "Oh my god! I love how they draw faces! But why don't mine look like that? What are they doing different!? I don't understand!" If you can identify the different categories of features and options available, you can spot what they're doing that differs from your own system.
  
  
• We can adjust our own styles more easily when we understand them at a conceptual level. For example, maybe you have a project where you need to rapidly produce very simple designs, but your brain is still stuck trial-and-erroring realistic proportions; it can't even begin to simplify the process down. That'd turn a simple low-scope project into a nightmare.

With that understanding of why these skills matter in mind, let's explore them in more depth as part of the designer's toolbox.

[holland]

But first, let's address the elephant in the room:

If designing is just selecting between preset options, why can't I have a generator do it for me?

Indeed, long before we had anything like ChatGPT, there were already countless programs and websites that could be used to generate designs.

And long before computers even existed, there were countless paper calculators, books of tables, and other mass produced devices that could be used to generate designs.

And long before there was the industrial revolution or printing press, there were countless tools and systems taught to generate designs through randomizers like cards, dice, clouds, and dreams.

So why do we even need designers?

Although the individual options may be multiple choice questions, the combinations chosen can be a bit more complicated. Like a story, a well-constructed design has some logic and theme to it. Things are pieced together how they are for a reason.

A variety of reasons, rather.

• On the production side, the designer concerns themselves with how reproducible that design actually is. For example, if you put every possible feature into a generator as an option and produce ten characters, those characters might be so radically different that each model needs to be completely rebuilt from scratch, or each illustration may be so laborious to accurately recreate those subtle variations in features that the project would go beyond over-budget.
  
  
• On the creative side, the designer concerns themselves with aesthetics, readability, and theme. Those characters from the previous example could be so different that they stick out from each other like a sore thumb and never look good when interacting. Or they could be generated with such similar features that players can't tell an enemy and an ally apart. Or they might just not make any sense in the world they're plopped into.
  
  
• On the ethics and marketing side, the designer concerns themselves with ensuring the designs properly communicate the experience the project is promising and attract the right crowds. For example, your generated antagonists might end up all being one race or one body type, which can give the impression the story is based around a bigoted concept that deters a whole community of viewers.

So even when experimenting with dice and such, you still need your designer's touch to correctly limit the options to only those relevant to the context, and you still need your designer's touch to correctly veto combinations that may be offensive, repetitive, clashing, etc.

There is also the issue of satisfaction. Both for the consumer and the creator.

• On the creator side, especially with solo projects, the design process might be something that's personally important to you. Maybe it's fun. Maybe it connects you to the characters and places. Maybe it makes you care about the project in a way that just following someone else's outline wouldn't.
  
  If you skip a step that would provide you a sense of satisfaction in your artistic process, you risk losing your momentum in a project or burning out completely.
  
  
• On the consumer side, their sense of connection to the humans that created something often plays a big part in the satisfaction they get out of it.
  
  They might find satisfaction in trying to figure out what the creator's thought process was. They might find satisfaction in seeing their own thoughts, feelings, and experiences reflected back at them from someone with a similar background making their creative design decisions from that context.
  
  Something being "lolsorandom" can offer its own satisfaction, but a lot of its appeal is lost when the source of the randomness is revealed to not even be a person, as the satisfaction often comes from wondering what kind of person would have made something so strange and why.

Remember that something being more efficient doesn't necessarily mean it's more effective. Rapidly generating designs -- whether through software or playing cards or brain games -- always risks losing some of the substance that makes both creators and consumers actually care about those designs.

I'm not saying you can't play with randomizers. I use divination techniques in my design work all the time. It's not cheating or anything. Just realize that over-relying on them without sufficient human intervention produces a more shallow experience for both you and your audience.

[holland]

Alright. So let's break down the basics in more detail and take a look into some of the tools we have in our toolbox as designers.

I'll be grouping these and submitting them in separate replies for your and my sanity, as it turned out to be longer and more involved than I expected it to be. Maybe this will be better suited to fully separate posts in the future...

If you've done any programming, these concepts might sound familiar. The way that data and objects are handled in software is one of the purest forms of design. Code really lays bare the fundamental nature of everything communicating across a program.

[holland]

Features in Design

A design feature is a data point that we would describe something with. For example, we might describe a piece of slate as being gray in color, flat in shape, and rough in texture. Those are its features.

Each feature is composed of a prompt (ex: "color") and its state (ex: "gray") at a bare minimum.

They may also incorporate options describing their potential states (ex: slate's "color" could also be "blue-gray," "green," "reddish," or "purple-gray") and defaults their state will defer to in the vast majority of cases (ex: we assume slate "color" is "gray" by default).

[holland]

Static vs Variable Features

Whenever we learn to design something new, the first thing that we want to do is identify its features and whether their states are static or variable. When we compare a collection of references, what stays the same between items, and what changes?

Variable features often change between designs, which requires the designer to make a decision about them. These become those multiple choice questions we've been discussing.

On the other hand, static features rarely change between designs, enabling the designer to gloss over them and focus on the variables instead.

For example, "number of toes" might be a static feature in a particular character designer's toolbox , while "length of toes" may be a variable one. Why? Because most people have 10 toes, and those 10 toes are shorter on some people and longer on others.

The variable balance between these two categories of features is very important:

If too many features are static, there will be a lack of variety in your designs. This can make it hard to tell designs apart from one another, such as to tell who spoke which line in a comic. And when a design does need to break the static default, it can feel like an offensive caricature when contrasted by a well-established monotony.

For example, if all your characters have button noses except for one Jewish individual, it reads as much more offensive than it would if the rest of the cast had a natural variety of nose types too. When you lock yourself into a mostly static model, you're also locking yourself out of communicating nuanced ideas.

But if too many features are variable -- or if the variable features have too many options -- the design process will become too overwhelming to complete and maintain, and there will be a lack of cohesion between designs. So we do need to set rather harsh limits when building our design palette for a project.

[holland]

Objective vs Contextual Palettes

A design palette is a combination of static and variable features used to design something.

Your objective design palette is the culmination of your studies. It encompasses everything you've learned and documented about the ways different design subjects may vary. But its breadth makes it increasingly unwieldy as you get deeper and deeper into your studies.

Your contextual design palettes are simpler and built to actually be used when designing in a particular context. They can be developed on the fly, appear as a side effect of the art style you're pursuing, or carefully curated over dozens of meetings in a corporate work environment.

For example, let's say that your references show chairs made of wood, stainless steel, bamboo, wicker, and plastic. In your objective palette, you recognize that these are all materials a chair could genuinely be made from.

But when designing chairs for a fantasy world where plastic does not exist and metal is very rare, you'd remove those two options from your contextual palette for chair designs in that world. You might also add material options unique to the world, like dragon bones.

Going forward, most mentions of design palettes in this post will be referring to contextual ones. The way objective ones are organized and used is a bit different.

[holland]

Variable Option Types

Although options are technically always a multiple choice question, if we think of them all as arbitrary lists of options to  be infinitely added to, we will quickly become overwhelmed.

Thankfully, most variable features actually have some logic behind them, instead of having to remember every individual option.

The first step to understanding this is identifying the various types and their options.

1. Toggle State Features only ever have two options, which boil down to yes/no, on/off, true/false, etc. This feature is basically just a question of whether it's present or not.

2. Spectral State Features exist on a range/spectrum between two extremes. For example, the height of a character might be somewhere between 0 and 9 feet tall. The feature's options can be described by these extremes of that range and visualized as a gradient or line graph.

3. Gamut State Features have multiple elements that each exist on separate but intersecting spectrums. For example, we can think of "color" as being composed of three spectrums: hue, value, and saturation. A gamut can be described by the combined ranges of its spectrums and visualized as a multi-dimensional form.

4. Typed State Features have a set of options that do not have a mathematical relationship to one another, but they all represent the same type of object / concept / thing. This typically presents as a kind of nested design. For example, when designing the outside of a building, you might select its "front door" feature from the types of outer doors you've designed or studied previously.

5. Assorted State Features have a set of options that are not clearly related to one another in any way. For example, the "motif" feature on a decorative pattern could be pretty much anything with a visual appearance. These are sometimes a smaller list of options, but usually not without an intentional constraint added.

6. Complex State Features are defined by a combination of states across a variety of different features. For example, the current state of a facial expression or body pose is much more complicated than a single state type can explain, and yet it is restricted by other simple features, like the range of how open the mouth can be. These end up requiring more abstract mental processing to reason through and manage.

[holland]

Palette Constraints

In their raw form, these features are useful at a conceptual level as part of our objective palette. But to make them easier to work with, we need to simplify them for our contextual palette. So let's look at the tools we have available for doing so.

One option that we have is to use a feature constraint. In this case, a feature -- whether static or variable -- is left completely absent in a contextual palette.

For example, when using an art style where all the characters have triangle noses, you have no reason to worry about all the different features of a realistic three-dimensional nose.

Another option we have is to use state constraints. In this case, we either make a feature static or restrict the number of options that can be selected between. This looks different depending on the variable type involved.

State constraints are more nuanced and complicated, so let's take a moment to explore them in detail.

[holland]

State Constraint Types

Freeze Constraint

A feature that would be variable is made static. It may use the original default for that feature, or a new default that's more appropriate for the context. For example, we might freeze the eye color feature when designing characters in a race where everyone has golden eyes.

Any type of variable state can be frozen.

Range Constraint

A feature that could be selected from any option on a spectrum is narrowed down to a limited range of that spectrum, by establishing a maximum and minimum value, or upper and lower limits.

This can also be applied to gamuts and complex states.

Although we touched on the issue earlier, I want to emphasize the importance of ranges, as it comes up constantly in design and is easy to slack on.

When you're studying a new feature that adjusts on a spectrum, you really want to find those maximums and minimums of what is physically possible. Otherwise, it's left floating around in your brain as this vague abstract concept with no real definition.

And when you're incorporating that feature into your process and style, you really want to narrow or expand that range to suit your personal tastes or what works with your own art in terms of tone, mood, realism, etc.

And then when you're dealing with more specific designs, you may need to narrow it further in context, or even expand it if a certain exception is required (like if you have occasional comedic tone-shifts).

Just be aware that having too many exceptions that break out of a nested range will shift the viewer's perception of what the range actually is. So it's no longer "the rules are x but this one exception is y," it's "the rules are y but a bunch of stuff is arbitrarily restricted to x."

For a common example with a complex variable, your model sheets might demonstrate the max/min for each expression your character makes, establishing their personal range.

Maybe the cool-headed loner never goes past shedding a single tear even at max sadness, while the comic relief can produce waterfalls from their eyes that defy gravity.

Despite both being human figures making facial expressions in response to the same emotions, their expressive range is different in the context of their personalities, within the context of their creator.

So your range isn't just what's physically possible, and it's not just what looks appealing in your style. There are all these nesting contexts where the ranges deeper in are restricted by the ranges of the higher context they exist within.

Scale Constraint

A state that could be selected from any point on a range is restricted to a limited scaling of that range. Basically, the level-of-detail (LOD) is lowered by some rule applied to the range.

Let's use character height as an example.

Being a length, you could measure height in any of these levels of detail, each of which represents its own potential scale:

- half nanometer
- nanometer
- half millimeter
- millimeter
- half centimeter
- centimeter
- half inch
- inch
- half foot
- foot
- half yard
- yard

...and so on.

There are no decimals here, except for the .5 allowed in the half scales.

Some of these scales are more useful than others. One of the ways we can parse that is by identifying their degrees of variance: how many options are they actually providing me?

For example, if we measure characters by the yard (= three feet, roughly a meter) within the minimum and maximum height range of, say 3 inches to 9 feet, this only gives us three degrees of variance: one yard, two yards, or three yards.

And when you look at where in the range those degrees are, they're not even very useful if your style is semi-realistic. 1 yard would allow us to depict toddlers, and 3 yards would allow us to depict unusual individuals with gigantism. 2 yards -- around six feet -- would be where the vast majority of our characters lie.

So for adults within a more common range of heights (5 feet to 7 feet), there is really only 1 degree of variance. That dramatically limits our design options and functionally freezes the character height feature.

Besides the degrees of variance, we also want to compare our scale to the margin of error in where and how our design will actually be applied.

For example, it is common in Europe and Asia to describe someone's height in centimeters, and Japanese comics will often list the exact number of centimeters or half-centimeters tall a character is in their design profile.

But be honest with yourself. If you're illustrating your characters without the help of a highly customized 3D model, it is very unlikely that you'll be able to reliably communicate a 1 centimeter height difference between two characters drawn posed in perspective.

Most skilled artists' margin of error when communicating lengths on big organic forms in perspective near the foreground is more like 1-3 inches.

So let's look back at that common height scale. 5 to 7 feet, a total range of 2 feet or 24 inches.

In centimeters, that would give us 60 degrees of variance in terms of design. Those are 60 different height classes we can place our characters in conceptually. This would be our design scale.

But with a margin of error in 3 inch increments, we can realistically only execute 8 degrees of variance within the common height range. So even if we design 60 height classes, we can only draw 8. This would be our applied scale.

Looking at it this way, those eight height classes do provide enough variety in terms of design for most purposes, and you could skip the more detailed scale altogether:



We'll talk about some funky stuff you can do to progress and convert between design scales and applied scales later. For now, I just want you to understand that scales exist in the first place as a tool to further simplify ranges.

I also want to point out that you have the option to remove pieces of the scale or subsets of the range as desired, or to shift parts of a scale slightly when the margin of error allows. Your scale doesn't have to be linear. It's a very subjective and preferential thing how exactly you opt to lay out your scales for your palettes.

Inherit Constraints

One feature's state is assumed from another feature's state. This is very common in proportional design systems.

For example, you might assume the distance between a character's eyes is equal to the width of one eye. Even though this isn't always accurate and limits the types of faces you can depict, it's a helpful simplification that usually results in a face that looks like a face.

So in styles where you don't need to get into the weeds of more specific proportions, you might apply this kind of constraint to that feature, giving you one less thing to think about when working.

You can think of this like a "living" freeze constraint. The state is adjusted automatically whenever the related features are adjusted, as opposed to being dependent on a particular design context.

Inherit constraints can be applied to any type of variable state.

Filter Constraints

One feature's state options are limited based on another feature's state, or multiple features' states. Unlike freeze and inherit constraints, filters may still require some design decision to be made if multiple options remain after filtering.

For example, you might have the feature "rate of atmospheric perspective" for your imagined landscape. You've already established the weather as rainy and hot, and that implies that there is steam fog and a lot of particles in the air that would increase the rate light scatters at. So you can already filter out the end of the spectrum associated with clear days.

But the most common types of filters are high-level concepts like motifs, themes / pillars, genres, and premises, amongst many others.

These constraints usually come out of other aspects of a project, such as its storytelling or the context it's being applied in.

A motif is any recurring element in a project or design. For example, you might have spiders, webs, chains, stripes, and coffins be recurring elements in a Halloween design. Then when you have a complex feature like the fabric pattern on a character's dress, you naturally have a narrower set of options to select between as opposed to "literally anything goes."

A theme or pillar is a motif that helps to tie the overarching design together at an emotional, narrative, or practical level. This gives it a sense of unity and helps to filter down and guide design decisions across the project as a whole. A project can have more than one.

In our previous example, you can think of "Halloween" as one of the themes tying all those motifs together.

For a more literary example, you might have the theme "innocence is defined by the cruelty of ignorance."

To communicate that, you then have your characters based around some ignorant past -- perhaps they are children, or from a sheltered home, or are dissociated from abuse, or have intellectual limitations -- that presents in some cruel way, like the sheltered rich kid looking down on the homeless.

And those might be contrasted by characters that express the other end of the theme's spectrum. So you have your characters based around some worldly past -- the social worker whose seen it all, the traveler who found humility in a culture unlike their own, a wealthy man who fell from grace and had to work his way back up from nothing, or the street rat that did the opposite -- and show their displays of knowing kindness towards others.

While, perhaps, the main antagonist character is someone on that end of the spectrum who instead uses their knowledge to abuse others, showing that while ignorance has a tendency towards cruelty, awareness only gives you the option to choose kindness in its place.

This theme would then filter out characters you think of with backstories and traits that contradict the core concepts you're communicating in a way that doesn't add any useful nuance to it.

Themes will be discussed more in the storytelling post, but be aware that they're a thing in most projects, even if they're not that deep. The main difference between designing something within and outside of a larger project is having to apply those thematic filters.

Outside a project or story or assignment, you have a lot more freedom, but freedom isn't always a good thing. Constraints force us to make more creative decisions, as we naturally try to push the boundaries in new and unusual ways.

A genre is an established set of motifs, themes, and needs that tend to occur in combination. The consumer is expecting at least some elements associated with a genre to be present if it's marketed as that  genre.

So establishing the genre of a project tends to filter your design options towards the expectations of that audience -- although it's good practice to throw some of your own unique ideas in, too.

Genres are not just a matter of fiction. For example, you might be designing a house, and to do so, you're expected to pick between established styles suited to different needs as your starting point when communicating with the client.

A premise or brief sets the initial scope and direction of the project, which naturally filters out many of the options outside of that scope. This premise is usually a combination of all the other high-concept filter types we've discussed here and is a more concise way to think about them.

You might have noticed by now that these filters are also implying a kind of feature constraint (not just state constraint) dependent on the stage of the design process that you're in... We'll dive deeper into that shortly.

[holland]

Feature Constraints

As a reminder, feature constraints deal with whether or not a feature is even on our radar as something that exists in the first place, whereas state constraints restrict the options for features that are present and considered when applying the design.

With that in mind, feature constraints are defined by the timing or scope they are applied to.

Scope Constraint

When a feature is determined to be out of the scope of a particular project or design, it is dropped altogether.

The example we gave earlier of a triangle nose style not requiring any nose form or anatomy features is an example of this type.

Process Constraint

When a feature is not relevant to a particular stage of the process, you may choose to temporarily leave it out.

For example, when you're establishing the initial premise of a story, it might not matter exactly how big of a gap there is between the main character's eyes, or what their favorite drink is, or even whether they are male, female, or intersex.

We'll talk more about these kinds of time-based progressions of design palettes in more depth later in this thread, as it's going to need a dedicated section.

Target Constraint

This is technically a type of process constraint, but I've chosen to pull it out here for emphasis.

Depending on what part of the project you're currently targeting as your focus, the features available to be considered may be different.

This broadly boils down to understanding the various things you're designing.

For example, it's very common for beginner writers to conflate project design, prose design, story design, environment design, and character design as all being the same thing, when these are entirely different areas of expertise.

Not only may you need to level up in them individually through more targeted training and thinking through their particular palettes, but you might be feeling like you're progressing on one part of the project when you're working on the another by not differentiating them.

You can spend years worldbuilding (a valid project of its own) thinking you're telling a story, only to realize you never established any plots within the world to actually write. I did this with a visual novel project once and have seen many others do the same.

This erratic targeting is sometimes just a byproduct of being distracted by the learning process.

You've misidentified a learning project that's way above your current level for a do-ing project that's below your level. So now you're beating yourself up for not meeting deadlines you never could have met, because you dramatically underestimated and misidentified the skills involved in the project.

Now you're getting caught up in the whirlwind of the learning process for a particular area that's becoming over-developed compared to the rest of the project. The brain is naturally trying to get as much practice in the new skill as it can to contextualize it, but that's negatively impacting everything else.

It's hard to get back on track when you've spent a long time off-target or over-targetted, and it's common for these projects to be set aside for many years while the skill gaps are addressed and a new perspective on the process is gained.

Filter Constraint

The filters discussed for states also have some relevance with features at a higher level, although how they apply to them is a bit different.

For example, when telling an upbeat fantasy story for children with themes around finding friendship, it might be unnecessary or inappropriate to consider the "severity of tax debt" feature of those characters' backstories.

In a way, all of the other types discussed here are also filter constraints of some kind.

Mostly, you just need to realize that not every feature/prompt is going to be thought about and used at every stage of the process, if it's ever used at all for that project or by that artist in particular.