Some time ago, our team was discussing how AIGC content might reach international audiences. A very practical question came up: if the same AI character reaches the United States, Mexico, and Japan, how should it change? The simplest answer, of course, is translation. Going further, you can change the dubbing, adapt the subtitles, or even adjust local turns of phrase. But as we talked, another question surfaced: why stop at translation? If this character was generated by AI in the first place, why not let the character actually live in Mexico once it arrives there? The streets could change, the room could change, the jokes could change, and even some stories could unfold again through the lens of local culture. Yet at the same time, the character would still be the same person. If the Episode One you see and the Episode One I see are no longer exactly the same, are we still watching the same work? And what if the future holds not three versions, but one hundred million?
For more than a century, we have been accustomed to one thing: once a work is finished, it is simply there. Once a book is printed, it is largely fixed. Once a film is edited, everyone watches the same film. Once a short video is published, a million people scrolling past see the same file. Algorithms change which piece of content you see; they do not truly change the content itself. Generative AI may be the first thing to loosen that premise: the algorithm may no longer decide only "what to show you"; it may also begin deciding "what this content becomes for you."
Recommendation algorithms offer a different version to each person: they select one item for you from a million pieces of content. Generative content could go a step further: first learn who you are, then generate this piece for you. The language, the city where a character lives, cultural context, humor, video length, narrative pace—even the story itself—could all change. The old model of tailored content for each person may gradually gain something new: a distinct work for each person.
But this immediately creates a problem: what is allowed to change? If everything can change, then the IP no longer exists. Suppose a character once made a deeply courageous choice to save someone. That experience has become part of the character. You cannot erase that past merely because a system discovers that another audience prefers a different kind of character. The real difficulty is not whether many versions can be generated, but what may change and what must never change.
Building generative IP in the future may require clearly defining its "variables" and "constants." The language can change. Clothing can change within a certain range. The city can change. Video length can change. Cultural expression can change. But a character's core personality, pivotal experiences, primary relationships, value-based choices, the foundational rules of the world, and facts already established in the story should not be changed lightly. The true boundary of generative IP is not what it can generate. It is what it explicitly prevents a generative system from altering at will.
This has also led me to rethink what we mean by "worldbuilding." In the future, worldbuilding will not only be about what a world looks like; it will also include the rules by which that world generates itself. Can a character return after death? Can time be reversed? Does the character know they are AI? Can users change the plot? To what extent? Without rules, the stronger the generative capability becomes, the more easily the world may fall apart.
A work may shift from being a "file" to being a "space of possibilities." Creators no longer prescribe the one fixed version every audience member must see. Instead, they define who the characters are, what the world is, what has already happened, what cannot change, and where change is allowed. Within those boundaries, the system then generates specific versions for different users, platforms, languages, and settings. In the past, creators decided "what you see." In the future, the creators of some works may decide "what you can see."
If stories can all be generated, what value do directors and screenwriters still have? I think the question is backwards. When there is only one version, a screenwriter writes one path. When there are ten thousand possible versions, the screenwriter instead needs to build a stronger story system. The same is true for directors. Once models can generate countless images, a director's increasingly important job may no longer be "make this shot for me," but "by what visual rules should this world exist?" Creators may gradually move from controlling every outcome to controlling the rules that produce outcomes.
What interests me most is not tailored content for each person, but whether a character might come to know you. Character Memory means a character remembers its own past. What if a future AI character not only remembers itself, but remembers you too? It remembers what you said when you first met, what happened between you, and whether you chose to help or refuse it the last time; it carries all of that into your next meeting. The past was "I know this character." The future may become "this character knows me too."
That also introduces another problem: who is the real them? An AI character may have one hundred million users, with each person having experienced something slightly different with it. Shared memory, personal memory, world memory, and the character's own memory may all coexist. Which memories become canon? Which belong only to this character and me? What happens when different users' experiences conflict? This is no longer only a question of content production; it is also becoming a question of system design and rights.
Copyright also becomes more complicated as a result. If a work is not a fixed file, but a set of rules plus a generative system, what exactly is the work? The character design, the worldbuilding, the model, the prompts, the generation rules, each specific video generated, or the entire system? If users participate in the plot, who owns that portion of the content? If a particular version contains infringing content, who bears responsibility? Establishing rights, granting permissions, keeping records, and maintaining a chain of responsibility will become increasingly important.
Globalization may be one of the earliest places where this form of content takes hold. In the past, when a Chinese IP went abroad, the most common approaches were translation, subtitles, dubbing, re-editing, and local distribution. But culture is not language. Generative AI gives us a possibility that was once expensive to pursue: rather than translating the same content for the whole world, allowing the same IP to develop adapted expressions within different cultures.
In the past, we sold one work to the entire world. In the future, we may let one world grow globally. This is not simply "work → global market"; it may become "world → continued growth across different markets." A character known to a user in Mexico and one known to a user in China may share a common past and core personality while also having some different experiences.
But I do not believe all content should become "one version for each person." Many great works matter precisely because everyone sees the same version. We have watched the same film together, remembered the same line, and gone through the same ending; those shared experiences are themselves part of culture. Fixed works and generative works may coexist for a long time.
What is truly worth studying is not one hundred million versions, but what remains unchanged after one hundred million versions. The same core character, the same history that cannot be casually erased, the same rules of the world, certain important relationships, the same value-based choices, and the same user understanding built up over time. Models are responsible for continually creating change; IP is responsible for remaining itself amid that change. In the past, content creation delivered an answer. In the future, some creators may deliver a set of rules, a character, a world, and a set of boundaries, then tell users: come in and see what happens between you and this world. When a work no longer has only one fixed version, what exactly is a creator creating? Perhaps the answer is no longer simply a work. It may be a world capable of continually producing works, relationships, and memories.
Copy this essay for someone thinking about AIGC content, IP, organisation or digital identity.
