joão (@fishesareweird) has been posting about how boring the AI image generation discourse is, with everyone optimizing for realism or engagement instead of seeing how far the thing can be pushed, and over the first days of October you can watch him find a style of his own. On the 2nd he posts four images that all came from the same prompt, which is just a wall of box-drawing characters with an ASCII fish in it:

The same day, the first clips of a low poly anime world, and a showreel of its animations and rendering:

On the 3rd it's the same characters in a clothing store, which became the first video on his new channel, real life tv:

And on the 11th he explains it. He writes the scripts, makes the 3D animations with Opus 5.5 and uses ElevenLabs for the narration, and four days in, the first video has 27.7k views, the channel has over 500 subscribers, and people in the comments are already invested in the lore of characters who never even talk:

This is a good example of creative AI usage, and it points at a nice idea: you can "own" an entire visual style like this and then release games with it. If joão decides to make a game in this style, he already has viewers who like it and who care about these characters, which is a nice path for marketing. Ideally you'd want something less generic than low poly anime, Kurzgesagt for instance has a very specific style that's very ownable, you see one frame of their videos and you know it's them. And of course people can copy a style immediately, so the moat doesn't last long if you do get popular, but it's a good idea if you haven't had a breakout hit yet.

The points below were added by Claude (Opus 5.5) on review of a327ex's initial idea, and he agrees with them:

What joão owns is a world, not just a look. The comments are invested in the lore of characters who never speak, and while anyone can copy low poly anime in an afternoon, nobody can copy his characters, his island and his tone. So the durable asset is the IP, with the style as its packaging: the style moat erodes, the world moat doesn't.

Copies can reinforce the original. If a style is distinctive enough, copies read as copies, people call them "Kurzgesagt-style", and that keeps naming the original. What erodes quickly is the novelty, not the original's status, so the window lasts longer than it seems, as long as you're the recognized reference rather than one of many who happened on the same look.

Video viewers aren't automatically game buyers. The transfer works best when the game is a natural extension of what they already like, a chance to spend time in that world with those characters. An unrelated game in the same style keeps much less of the audience, so the game should be designed around the world, not just reuse the look.

It fits the view from #172 that marketing changes the speed, not the ceiling. An audience built in advance doesn't make the game better, it means launch day starts with people who already care, which matters most on Steam, where a strong launch feeds visibility. That's why it's most valuable before a breakout hit: after one, your name does the job.

And the AI part is output rate. Making a style is cheap now, so styles will multiply and novelty alone won't carry anyone. What's hard in what joão is doing is volume and consistency, two videos in four days, same world, same characters, and AI is what makes that cadence sustainable for one person. The cadence is what builds the audience.