Interactive Narrative

(ME): "are there any projects/research about: "ai to put yourself into a movie"
like, ai re-enacts a movie script and you take over the role of a character and if you make changes, the plot changes? or reenacting in VR like in ready player one?"
(CHATGPT:)
Yes. The field is usually described as LLM-based interactive drama, generative interactive narrative, playable television, or interactive generative video.

No polished system yet fully combines:

1. an existing feature film;

2. you physically or vocally performing one character;

3. AI actors performing everyone else;

4. unrestricted deviations from the script;

5. a coherent newly generated plot; and

6. real-time cinematic VR.

Several projects are surprisingly close to individual parts of that idea.

Closest direct research

1. From Role-Play to Drama-Interaction

This is probably the closest academic match to what you described. The player enters a known story, assumes the role of one character, speaks or takes actions, and interacts with characters controlled by an LLM.

The researchers built demonstrations based on Detective Conan, Harry Potter, and Romeo and Juliet.

Their “Narrative Chain” system tries to keep the original dramatic progression intact while still responding to the player—a central problem in your idea: what happens when Harry refuses to go somewhere, warns someone early, or otherwise breaks the screenplay?

• From Role-Play to Drama-Interaction: An LLM Solution

A 2025 follow-up studies:

• immersion: feeling present inside the story;

• agency: genuinely influencing the story rather than merely choosing dialogue;

• adapting NPC reactions to the player’s intentions while retaining dramatic structure.

• Follow-up paper on immersion and agency in drama interaction

These experiments are primarily dialogue-driven rather than photorealistic movies or VR, but conceptually they are almost exactly your proposal.

2. IBSEN

IBSEN divides the AI into a director agent and multiple actor agents.

The director maintains the overall plot, while the actor agents portray individual characters. When a human enters and changes events, the director can revise the plan and redirect the actors toward a new dramatic objective.

That “AI director repairs the movie after you disrupt it” architecture is likely how a practical version would work.

• IBSEN: Director-Actor Agent Collaboration for Interactive Narrative

3. StoryVerse and WhatELSE

Autodesk Research’s StoryVerse generates “living stories” from high-level plot outlines, character simulations, and the changing state of the story world.

The plot is co-created by the author, simulated characters, and player rather than selected from a fixed branching tree.

• StoryVerse

Its successor, WhatELSE, lets an author establish a range of permissible alternate plots.

For example, a studio might decide:

“The detective may accuse different suspects, lose evidence, or become allied with the villain—but the story must remain a murder mystery and eventually reveal the conspiracy.”

The system then produces playable events inside those boundaries.

This is particularly relevant to adapting films because it provides something between a completely fixed script and unrestricted AI improvisation.

• WhatELSE

4. NarrativePlay and Open-Theatre

NarrativePlay converts existing narratives into environments where the user selects a fictional character, explores the story, and converses with AI-controlled characters.

It extracts personalities and major events from the source narrative and adds generated settings and character portraits.

• NarrativePlay: Interactive Narrative Understanding

Open-Theatre, published in 2025, is an open-source toolkit specifically for constructing LLM-based interactive dramas.

It is a more practical starting point for experimentation than most research-only systems.

• Open-Theatre: An Open-Source Toolkit for LLM-Based Interactive Drama

5. STAGE

The 2026 STAGE benchmark focuses on whether AI can understand entire movie screenplays well enough to construct character relationships, track events, and perform in-script character role-playing.

It contains structured material covering 150 English- and Chinese-language films.

It is not itself an interactive movie, but it addresses a prerequisite: the AI needs a reliable internal model of what everyone knows, where they are, what has happened, and how each character should behave at every point in the film.

• STAGE: Screenplay Understanding and Character Role-Playing

The closest actual consumer product: Showrunner

Fable Studio’s Showrunner is currently one of the most direct visual implementations.

Users can:

• create or remix animated television scenes;

• upload themselves or friends as characters;

• place those characters into a show’s world;

• provide instructions for what happens next;

• generate scenes or episodes rather than merely receiving text.

The app describes itself as serialized AI storytelling and explicitly offers the ability to “put yourself in the show.”

• Showrunner app

• Showrunner Agents research

Showrunner is not yet “act continuously inside the movie.” You prompt or remix a scene and then wait for it to be generated.

It currently works best for relatively self-contained animated episodes, especially formats where characters can reset between installments, rather than a causally complex two-hour movie that continually branches.

Its creators nevertheless frame the eventual concept as making films and television playable, including putting yourself into an existing fictional world and generating alternate continuations.

The visual technology is beginning to catch up

Runway’s 2026 Characters system can generate a responsive, expressive video character in real time from a single reference image.

A language model and voice system control the conversation while Runway generates the character’s visible performance.

That could eventually supply the AI “actors” in this kind of experience.

• Runway Characters

Runway’s Game Worlds research is more directly aimed at nonlinear experiences where users explore characters, stories, and environments generated with AI.

It is still an experimental hybrid rather than a convincing cinematic world.

• Runway Game Worlds

CinemaWorld, a 2026 mixed-reality research system, extracts objects and effects from a film and extends them spatially into the viewer’s physical surroundings on a Quest 3.

It does not alter the story, but it shows how an ordinary 2D movie could be converted into a spatial experience.

• CinemaWorld

What is still missing

The main obstacle is no longer merely generating dialogue. A complete system must simultaneously maintain:

• Narrative causality: If you prevent an early death, every later scene dependent on that death must be reconsidered.

• Character knowledge: Someone cannot react to information they have not learned in the altered timeline.

• Character arcs: Characters should evolve according to what has happened rather than act like static chatbots.

• Long-term visual continuity: Faces, clothing, rooms, props, injuries, and spatial arrangements must remain consistent.

• Real-time generation: VR generally cannot pause several minutes whenever you speak.

• Dramatic quality: Total freedom often produces shapeless role-play rather than something satisfying as a movie.

• Rights and consent: Commercially inserting users into recognizable films would require licenses covering the story world, characters, performances, voices, likenesses, and generated variations.

Researchers in interactive generative video still identify real-time rendering, persistent memory, long-horizon coherence, physics, and causal reasoning as major unsolved problems.

• Interactive Generative Video survey

What a plausible near-term version would look like

The first convincing version probably would not generate photorealistic video from scratch.

It would likely use:

1. a licensed movie screenplay converted into a scene and causal graph;

2. a game-engine reconstruction of the sets;

3. AI-controlled character agents;

4. an AI “director” that maintains dramatic goals;

5. voice input and VR body tracking for the player;

6. fixed or motion-captured animation augmented by generative facial performance;

7. dynamic dialogue and moderately flexible alternate scenes.

In other words, it would initially resemble immersive theater inside a video game, not a perfectly regenerated Hollywood movie.

Fully generated video could gradually replace the game-engine layer as interactive world models become more stable.

The research phrase most likely to produce useful results is:

“LLM-based interactive drama with player agency and embodied virtual characters”

Closely related searches include:

• generative drama management

• interactive narrative planning

• story-world simulation

• playable television

• interactive generative video