So, today, a content creator announced that they generated 50 emotes with ChatGPT in three minutes, acknowledging the stigma around AI but arguing that the results were great. The post got a lot of engagement, largely because of people objecting AI and because of how bad all the emotes looked… but it’s rage bait. It doesn’t matter to the original creator whether or not the 3-minute-emotes look good or not since they were merely looking for their 50 seconds of fame. They got that. So, we’re not talking about that.
I wanted to instead take the opportunity to talk about why generative AI cannot replicate a good emote and what actually makes an emote a good emote. I think it’s more important to understand why genAI looks the way it is than to point out that it looks bad and uncanny in these instances. People who pay for Twitter get money through ragebait, after all.
I hope that this post helps folks understand the matter a bit easier and encourages folks to either learn how to draw themselves or to reach out to human artists instead.
What actually makes an emote good?
Emotes aren’t just a piece of artwork that you stick into chat but rather a means of communicating something. Think of simple emoticons and how they drastically transform a sentence’s meaning.
“I’ve eaten my food :)” and “I’ve eaten my food :(” mean two very different things, after all.
As such, emotes on Twitch, YouTube, Kick, etc. are used to convey a message in far less characters, signal a belonging to a community, communicate an inside joke, and do many more things… but to do so, it needs to actually look good. To answer the question as to what makes an emote “good”, we need to first look at a few different factors that I personally pay a lot of attention to when I draw my own emotes.
Readability
For starters, emotes need to be visible and recognisable no matter what device configuration a viewer is using. Depending on the contrast of its colours, for instance, it gets difficult to read an emote when a person uses dark mode or light mode. Using an outline can sometimes fix this but generally speaking, artists need to adjust colours to ensure readability on all devices.
Similarly, mobile devices shrink the already small emotes down to an even smaller size, resulting in situations where emotes look like tiny dots if one doesn’t exaggerate their look. Rather than forcing all of a character to be visible in the emote, one needs to figure out how to position them and the important bits (gestures, expressions, and recognizable elements) in the frame without losing the connection to the mascot or character in question.
Immediate communication
The viewer shouldn’t need to study it to understand the intended emotion. A good emote doesn’t take more than a couple seconds to understand what it wants to convey.
For this, a lot of artists rely on stylistic devices found in media like cartoons, anime and games. People will understand flames or popping veins (💢) as something that means anger, whereas tears symbolise discomfort or sadness. Depending on the context, an emote is used for many different things, too, which is why artists can experiment with the composition and introduction of such elements.
Exaggeration
Expressions and gestures generally need to be exaggerated because subtlety gets lost when shrunk down. Generally speaking, when I draw emotes, the most important parts for me are the face and the hands (if present)… so I naturally make those bigger even if the proportions are incorrect. A “praying” emote would need to have a much bigger focus on the hands, after all, than the expression of a character… otherwise it might just look like a character is sleeping or resting their eyes.
Distinctiveness
A set shouldn’t contain ten slightly different versions of the same expression. An emote can represent an inside joke, recurring stream moment, character trait, or community meme. There are a lot of things that can be turned into emotes, so if two emotes can be used in the same context, it’s generally speaking better to replace one of them with something else.
“But the AI-generated emotes look good”
Now, some people will say that AI-generated emotes look good, and I will partially agree. They’re trained on human data, after all, so AI-generated artwork can look alright… because of its datasets.
AI can produce:
- appealing characters,
- clean compositions,
- expressive faces,
- attractive colors,
- technically impressive illustrations.
And if someone looks at one at full resolution and says “That looks pretty good,” they’re not necessarily wrong… but this also doesn’t necessarily mean that these are good emotes.
AI can replicate artworks sort of well but it generally speaking doesn’t understand what makes a good emote. Emotes need to fulfill a lot of criteria to be presentable… but once shrunk down, even a fully rendered artwork will look pretty mid because it gets too busy to discern.
The main issue here is that AI doesn’t actually draw emotes or think about what it’s creating. It merely predicts what you want based on what you’ve given it as a prompt. The aforementioned streamer got a ton of views by posting their AI-generated emotes, so another person copied their tweet word for word and posted more emotes generated by ChatGPT… and they looked exactly the same as the first batch of emotes.
On top of not understanding the subject matter of what makes an emote good, AI will also frequently (still to this day) not understand what makes expressions good or bad. The ones prompted by that one creator all have either a closed or an open mouth, but the shapes are all the same, making it hard to understand what difference there are between emote number 24 and emote number 49 or whatever. Yes, the streamer got “50+ emotes in three minutes” but they all look very similar and don’t fulfill their purpose. They’re not good emotes.
Imitation versus Design
Generative AI has been trained on enormous amounts of human-created material. Naturally, it can reproduce many of the visual conventions associated with emotes. But reproducing those conventions isn’t necessarily the same thing as understanding why those conventions exist.
As an example, I draw my own emotes and have done so since 2020. At first, I used to shade with black and highlight using white. This posed huge issues as the emotes became hard to read in both dark mode and light mode. I eventually moved on to shading using a colour but darker, which was tedious and kinda worked, but ultimately, I moved on from that and decided to use a low opacity multiply-layer that I’d colour in using red or blue based on what I need.
The reason for this is simple: When you use warm colours it conveys other emotions than when you use cold colours in shading.
AI also does shading, I guess, but it doesn’t give it much “thought”… Evident by all the emotes looking the same and depicting the same character, but the emotes conveying happiness and the emotes conveying sadness had the same type of shading.
Now, you don’t have to understand why certain colours invoke certain feelings to get what an emote or illustration tries to convey when you look at the colours. The absence of that, however, is what differentiates generative AI from human-made art. Humans design artworks by sketching dozens of times, revising a couple, going back and forth between different versions, and ultimately settling on a final composition. AI merely imitates what commonly is perceived as good-looking. If you ask a human to draw a clock, they’ll just draw a circle with two lines on it. LLMs will, however, generate a specific time displayed on the clock because 10:10 is seen as the most aesthetically pleasing time to display.
Obviously, humans take way longer to create things based on their skills, availability and how much they charge, but if you want to have good emotes, it’s probably good idea to get human input, at least part of the way, no?
AI makes generating possibilities cheap. It doesn’t automatically make choosing the right possibility easy, however, because people are easy to trick into liking the first result that comes up.
Conclusion
I think it’s important to distinguish between good designs and convenience. If you just need to fill slots, why not just draw something yourself, right?
I follow streamers who have literal stick-figures for their emotes, and they’re way more distinct and expressive than anything that was generated by that creator’s ChatGPT prompt.
Of course, human artists take longer… but you also get to work with someone who knows how to make emotes readable, how to get the mascot character to be recognised, and how to make an emote look good despite the need for simplicity.
Now, obviously, this is also a price question. A lot of AI bros will go out of their way to complain about artist pricing and revisions and all of that but in reality, you pay for an artist’s input, their experience, and their time. Of course, an artist could just draw what you want them to draw but if that doesn’t end up looking good, you’d blame the artist. As such, artists will insert their expertise into the process and help you come up with something, and to that end, they’re happy to make revisions happen since extra labour is better than an unhappy customer.
But also, as I mentioned already, I draw my own emotes and they looked pretty bad when I started in 2020… and now I wouldn’t want anyone else to do them for me because of how proud I am of most of them… and the ones that still stick around that don’t look too great are soon going to be replaced by other emotes created by me. It’s something I can take pride in. I’m happy about improving at a skill so much and sticking to it.
Of course, if you don’t want to do that, you can also ask friends to draw stuff for it, which adds a personal note, or you look for new artists looking to increase their portfolios… or you reach for your wallet and go on Etsy or VGen.
There’s a lot of options, and picking humans to do the job is always the better option.
This post was originally published on Indiecator.
If you wanna get in touch with me, feel free to stop by for one of my Twitch streams. For business inquiries, please reach out via email.
Indiecator 
Leave a comment