If Game Art is Feedback; GenAIArt is Whiteboarding
- Nov 13, 2025
- 6 min read
Updated: Dec 17, 2025
Game Art as Gameplay Feedback
The idea that Art exists in a game to help guide gameplay.
You've probably heard a picture is worth 1000 words.
But have you also heard of the sentence that changes meaning based on which word you emphasize?
I didn't say he stole my money
I didn't say he stole my money
I didn't say he stole my money
I didn't say he stole my money
I didn't say he stole my money
I didn't say he stole my money
I didn't say he stole my money
When Game Artists talk about Feedback, they're talking about using Art Skills on the 1000 words of a screenshot of the game, to place emphasis on the words that tell the player important things about playing.
This is recognized by Players in a straightforward Action-Reaction kind of way (ex: approaching an enemy gains aggro). But it's only recogniz-able because behind the scenes Artists have constructed an entire Visual Hierarchy so that emphasis - and by extension the player's attention - can shift when what's important changes from one moment to the next.
I didn't say he stole my money
...makes sense not just because stole is emphasized, but because all the other words aren't. Stole is being given the room it needs to stand out, like how good Visual Hierarchy gives Feedback the room it needs to stand out.
Art is used to make Feedback in games, because it takes 16 milliseconds to see an enemy looking at you, and nearly 2 whole seconds to read a sentence about aggro. That speed is important, because maintaining the entertained mental state of being 'absorbed' in a game is dependent on having speedy feedback.
Feedback involves responses to player actions; like low health warnings, communicating core mechanics, stopping the player from advancing beyond the level boundaries, etc. Hierarchy encompasses that and also things we don't really think about (because that's the point); where valid vs. invalid path is, when something can be picked up vs. when it's ambient clutter, making sure things in the distance aren't distractingly detailed, etc.
Why GenAIArt struggles: It doesn't turn down the Radio to see the Road better
GenAIArt's inability to factor in Human Perception is a problem for Feedback and Hierarchy.
The goal of Art isn't to draw a picture, it's to draw with the intent of manipulating how a Human perceives that picture. That manipulation doesn't always neatly coincide with the pixels in a picture, much less GenAIArt's internet common consensus of what those pixels should be.
For example: There's a moment when you're playing a game where HUDs "disappear".
They don't literally disappear, it's just that we Humans stop actively perceiving them because they're not giving us any new information.
A Human making a HUD understands this, and tries to make it visually unobtrusive, so it "disappears" easily. To GenAI the HUD never "disappears", let alone "disappears" in a way that's better or worse than other HUDs.
So why does that matter? GenAI may not be able to factor in Human Perception directly, but surely if it trains on enough unobtrusive HUDs, it'll repeat the unobtrusive HUD patterns...right? Well, not quite.
GenAI trains by creating surface-level summarizations of how things connect that tends to gloss over nuance. Multiple iterations of lost nuance engreebles the result in way that doesn't get reigned in by a Human Perception sanity check.
Ex: Minimaps from CS:GO, Read Dead Redemption 2, and Halo4:
(Unobtrusive Minimaps that communicate lots of information at a glance: direction player is facing, enemies are facing, walkable area, location of allies, enemies, enemies off-screen, allies off-screen, objectives, directions, and/or enemy type)
Ex: Gen AI Minimaps:
(Engreeblement makes these Minimaps unnecessarily detailed running the risk of too much emphasis, while conveying a fraction of the information above.)
This also leads to tangentially-related concepts getting entangled, because that lost nuance really helped to distinguish between two very different things.
So GenAIArt will blindly erase the unobtrusive HUD patterns, via accidentally exaggerating past them, and glossing over the important parts that make a pattern what it is and determine what it's for.
It's not like folks haven't tried to put normal human oversight in GenAI training. But for a number of reasons both human and math; it ends up less of a hard specification, and more of a gentle suggestion with fingers crossed. Good for an informal broad sense of style, not so much for absolutely guaranteeing results conform to Human Perception. If anything, it seems to open up opportunities for problems to compound. (I'm guessing because the oversight strategy is based on 'likability' instead of art fundamentals, which creates a "just because it's popular, doesn't mean it's useful" problem.)
Ex: "Futuristic technologically advanced solarpunk planet, highly detailed, temples on the clouds, one massive perfect sphere, bright sun magic hour, digital painting, hard edges, concept art, sharp focus, illustration, 8k highly detailed, ray traced" Prompt on Base Trained Model :
Prompt after Humans answered questions that were used in math to fine-tune :
Where I'm going with this, is that most images GenAIArt generates are like:
I didn't say he stole my money I didn't say he stole my money I didn't say he stole my money
Each word is clearly recognizable as a valid word, so you can read the sentence, but it's what it's ultimately trying to tell you is unclear. Because the writer is predisposed to too much emphasis (engreeblement), doesn't understand what separates one word from another (entanglement), and can't correct itself because it can't read (can't perceive).
If you believe Game Art exists to help guide gameplay, you have to come to terms with the fact a summary of an aggregate with a bias to bombast and no sense of human limits is bound to generate overwhelming, distracting, or confusing results.
The whole point of Feedback and Hierarchy is to create imagery that is clear, intuitive, and instructive to the Player. So you end up in the position of having to painstakingly reconcile all the ways it fails to be Human.
"If you want people to communicate something, make it easy"
Instead I'd suggest reframing GenAIArt as hyper accessible whiteboard doodling. Use GenAIArt to lower the barrier of entry for Non-Artists to collaborate with Artists on a visual idea before Concept Art is made.
This is particularly valuable for Feedback and Hierarchy, because it's a very visual topic that's hard to cover in a purely verbal way, and also a massive coordinated effort that inherently involves different disciplines. (ex: Game Designers tell Artists what it's important to have feedback of)
Disciplines who can struggle to follow along and translate their thoughts when the topic turns visual. So much so, they can be reluctant to discuss, which can result in spinning wheels on concept art shots-in-the-dark. Art-lead meetings can have a surprising blind spot about this, because Artists tend to underestimate how good they are at visualizing compared to your average person.
Using GenAIArt like this takes advantage of it's strengths (speed, accessibility), and minimizes the impact of it's weaknesses (Can't take Technical Limits or Art Rules into consideration, but that's what Concept Art is for).
Use a GenAIArt Tool in early meetings between Artists and Non-Artists. I'd suggest Tools that offer these features as options:
"In-Painting" (revision that suits the non-linear thinking of early brainstorming)
"Text-to-Image Generation" (For the Non-Artists)
T2I Adapter "Sketch-to-Image Generation" (For the Artists)
Ability to paint directly (anyone whose comfortable whiteboard sketching)
Segregation of Results on the same image (Layers, Regions, Inf. Canvas, etc.)
This is based on the Rule of Thumb 'Nudge' from Behavioral Economics, "if you want people to do something, make it easy".
This is about communication, not ideation. If you don't walk into meetings with some idea-ingredients to throw in the Think Soup, everybody's going to walk out of the room with a bland taste in their mouth. GenAIArt is too boring to be a good Creative Assistant, but it's great as a Drawing Assistant.
References:
Dissidia picture:
Troxler's Fading: https://en.wikipedia.org/wiki/Troxler%27s_fading Spheres Picture:
clock:
Astronomical ring dial https://curiousminds.co.uk/products/hemisferium-miniature-astronomical-ring-dial?_pos=2&_sid=685d033cc&_ss=r
Gamefootage of HUDs:
https://www.youtube.com/@MKIceAndFire
https://www.youtube.com/watch?v=k1VW8znw7Uc