AI and Virtual Romance: How ChatGPT Is Changing Fandom, for Better and Worse

AI is no longer a technology for a handful of specialists. It's becoming an everyday tool for making images, polishing text and collecting information about your oshi. In a survey centered on university students, 66.4% of respondents were reported to use generative AI every day.

The "Ghibli-style" image generation boom of spring 2025 is an example of an experience spreading, that anyone could transform their own photo, before anyone had explained the technology. Sam Altman changing his profile picture became part of the buzz too.

Before trying it, fans want to know more than whether it produces a fun image. If they can tell whether an uploaded photo is stored, whether they can choose who sees it, and whether it's fine to feed in an image of the oshi themselves, they can judge for themselves where play ends and privacy begins. Here's a basic point worth checking now that these tools are part of daily life: a photo showing friends or other people at a venue can't be used on one person's judgment alone.

Every time a new technology appears, the same debate repeats. Photography, recording and synthesized voices were all called fake at first.

Things usually settle into a division of use. Redrawing the line between what people do and what they leave to a tool, people have found a way to meet in the middle.

A friendly reply is a different thing from the other side actually intending to understand you. Even when you use it for fun, check whether conversations are saved, whether they're used for training and what happens when you cancel. If the time or the cost becomes a burden, think of it as a relationship you're free to step away from for a while.

Image generation moves fans who only watched over to the making side. Combine your own room, your oshi color and a favorite scene, post it to social media, and you're giving back an interpretation instead of just receiving a world.

Because you can describe an image in words and keep experimenting, even people who aren't good at drawing can become makers. The wider the door to participation opens, the more the official side needs to make the scope of secondary use easy to understand.

If the official side provides oshi colors and background materials, for example, people can join the world without editing a photo of a person at all. Instead of just lining up sample works, a company can keep beginners from crossing a line unknowingly by putting usable materials, expressions to avoid and required labels for posting on the same screen. Showing input examples and pre-publication checks alongside finished images also narrows the gap in creative experience.

Oshi-katsu is also a field that adopts new tools quickly. A practical attitude of using whatever works has taken root.

The gap between generations isn't really that big. It's less that young people have no resistance than that most people judge by the situation and the purpose of use.

Using it as a handy tool and using it in place of the person are different things. As long as that distinction is shared, conflict is unlikely.

Shingo Murakami of the idol group SUPER EIGHT is trying out appearing alongside an AI avatar, putting the person and an artificial double on the same screen. How the person's voice and character are handled is the line between fun and trust.

An AI avatar can run at any time, but keeping the parts the person supervised separate from the automatically generated ones keeps fans from getting the wrong idea.

The more the voice and replies sound like the person, the more easily fans take them as real statements. If AI use is labeled per stream or per post, and important announcements and apologies are decided to be the responsibility of the person or the team, a convenient double won't overwrite the person's will. Settling even how to correct a reply that differs from past statements saves fans from having to work out what's true.

Honestly, the easier participation is a good thing. People can now join in the expression even if they can't draw or don't have the equipment.

At the same time, we need to think about the people who spent years building their skills. New tools always shift the position of existing makers.

In SANYO's "AI Miss Marin-chan," fans vote to pick a grand prize winner from 12 candidates created with image generation AI. You don't just root for a finished character; you take part in the selection.

When it's clear who decides the AI candidates' settings and growth, and how the voting period and duplicate votes are handled, the fans' choices live on as a story.

Participants will also want to confirm whether the conditions are the same for every candidate and how proposals that weren't chosen are handled. By showing the evaluation criteria and what will be reflected before voting, and reporting back after the result on which opinions were adopted, a company can make voting a joint production rather than a number for promotion. The party responsible for the selection should be named, too.

It also matters who can use, after the result, the names and settings that emerged while fans were rooting for the candidates. If the scope in which fan ideas are adopted, the treatment of credit, and whether rejected ideas are kept or deleted are shown in advance, participants can vote knowing where their own creations will end up. A deadline for withdrawing a post is needed as well.

The debate about AI is hard because talk about technology and talk about feelings get mixed together. Whether something is technically possible and whether fans can accept it are separate questions.

Technology that recreates the voice of someone who has died, for example, is already at a practical stage. Whether to use it comes down to a judgment about how to treat the wishes of the bereaved and of the person, not whether it can be done.

It's no different when the person is alive. How far does the contract grant permission, and can it be withdrawn later? Pressing on while that stays vague leads to big problems afterward.

Fans are asked to use judgment too. Enjoying it as a derivative work is the same as before, but putting it out in a form that looks like the person's own statement invites misunderstanding even when that wasn't the intent.

An avatar the person plays and a derivative work fans make are not the same thing. Trust builds when the official side shows labels saying whether something is supervised by the person or is a fan's interpretation, and doesn't blur the boundary.

Likeness, voice, an artwork's style and training data raise questions of rights and ethics. People can enjoy it with peace of mind only when the person's permission, the scope of use and even the method for takedown requests are in place.

Fans also need somewhere to take the discomfort they feel. Take problems such as statements by someone other than the person appearing to be the person's own, or notifications continuing when you want to stop: if fans can choose to fix the labeling, delete their history or stop using the service, they can adjust how close the relationship is themselves. Companies shouldn't measure success only by usage time and reaction counts; whether people were able to leave and whether inquiries could be answered are also measures of operational quality.

In oshi-katsu, AI is already used as a tool: translation, summaries, rough image drafts. It works every day in places that don't stand out.

The only part that gets debated is what involves the person. If everything is treated as one problem, the discussion goes nowhere.

An AI oshi is convenient: it responds whenever you like. But when the distance gets too close, it can also become a design that consumes the user's time and money excessively. Fans need an experience that includes the freedom to take a break.

There's no need to replace with AI the chance failures and growth that only a human oshi has. Dividing the roles lasts longer: AI widens the range of creation and participation, and people carry the responsibility and the relationship that make up the story.

The quality of AI participation can't be measured only by how many works were generated. If people can tell person-supervised work from fan-made work, know how the material they entered is handled, and choose to delete or leave when something feels off, that reversibility lets them enjoy the convenience while protecting the wishes of the person and the fans.

Technology moves fast, and rules decided now soon go out of date. That's exactly why sharing criteria for judgment is more practical than detailed prohibitions.

Does the person agree, and might the people receiving it misunderstand? Just checking these two things each time would avoid many problems.