Storytelling

From PRIMO.ai
Jump to navigation Jump to search

YouTube ... Quora ...Google search ...Google News ...Bing News


Storytelling is one of the oldest ways humans share knowledge, build trust, and make sense of experience, and it has become one of the most visible frontiers of artificial intelligence (AI). In AI the subject runs in two directions. AI systems now act as storytellers and co-authors, generating plots, dialogue, images, voices, and video through generative AI, while storytelling also serves as a tool for making AI itself understandable, from narrative explanations of model decisions to the user stories and personas that shape product design. Both directions matter for education, marketing, entertainment, journalism, and everyday human-AI interaction, and both are changing quickly as large language models and video generators improve.

Computational storytelling is not new. Researchers built rule-based story generators in the 1970s, interactive drama systems in the 2000s, and neural story models in the late 2010s, long before chatbots made story generation a mass-market activity. What distinguishes current systems from conventional authorship is that they produce narratives probabilistically, at enormous scale, and in response to each reader or player: a single model can draft a bedtime story, voice a game character, or adapt a plot to a learner's choices in seconds. That flexibility comes with real limitations, including weak long-range plotting, a tendency toward predictable or overly positive arcs, unresolved copyright questions, and evidence that widespread AI assistance can make creative work more alike.

Part of the reason storytelling attracts so much AI research is that narrative appears to be central to human cognition. In a 2010 study published in the Proceedings of the National Academy of Sciences, Princeton researchers used fMRI to record a speaker telling a story and the listeners hearing it, and found that the listeners' brain activity became coupled to the speaker's, with stronger anticipatory coupling predicting better comprehension and the coupling disappearing when communication failed. Findings like these, explored further under neuroscience, help explain why designers, educators, and marketers treat stories as an unusually effective way to transfer ideas, and why AI systems that tell stories well could become persuasive and influential in both helpful and harmful ways.


Stories are how people make sense of experience, and AI now takes part in telling them: the most useful systems treat the machine as a collaborator that drafts, adapts, and explains, while humans supply intent, judgment, and meaning.


A Brief History of Computational Storytelling

The first influential story generator was TALE-SPIN, written by James Meehan at Yale and described in a 1976 research report supported in part by the Advanced Research Projects Agency. TALE-SPIN produced Aesop-style fables by simulating characters with goals, relationships, and knowledge about the world, then reporting how they planned and acted; when goals were poorly chosen or the story world did not support them, the result was a short or oddly structured "mis-spun" tale. Later systems such as Universe (Lebowitz, 1984 to 1985), MINSTREL, BRUTUS, and MEXICA explored author goals, case-based reasoning, and cycles of engagement and reflection, all within the tradition of symbolic AI, where knowledge and plans are written out explicitly rather than learned from data.

A second tradition focused on interactive narrative, where the challenge is keeping a story coherent while a human participant acts freely. Carnegie Mellon's Oz Project represented characters as autonomous, embodied agents, and later systems added a drama manager, a coordinating component that steers character behavior toward interesting, well-structured experiences without taking control away from the player. Michael Mateas and Andrew Stern's Façade (released in 2005) became the best-known example, and researchers subsequently framed drama management as a search and optimization problem, including experiments with Monte Carlo tree search and reinforcement learning. Interactive fiction author Emily Short showed with the Versu system how AI-driven characters choosing among pre-authored options could produce stories that differ on every playthrough.

Neural approaches arrived in the 2010s with recurrent neural networks and sequence-to-sequence models that learned storytelling patterns directly from text. A landmark 2018 ACL paper by Angela Fan, Mike Lewis, and Yann Dauphin of Facebook AI Research introduced the WritingPrompts dataset of roughly 300,000 human-written stories paired with prompts from a Reddit forum, and showed that hierarchical generation, first producing a premise and then expanding it into a story with the help of a gated multi-scale self-attention mechanism, improved coherence; human judges preferred the hierarchical model's stories about two to one over a strong non-hierarchical baseline. The arrival of GPT-2 in 2019 and later GPT models moved story generation from research labs to the public, first through games such as AI Dungeon and then, after ChatGPT launched in November 2022, through general-purpose assistants used by millions of writers, teachers, and hobbyists.

How AI Generates Stories

Language Models as Storytellers

Modern text-based storytelling relies on large language models built on the Transformer architecture. These models are autoregressive: at each step they use attention mechanisms to weigh the earlier tokens in their context, compute a probability distribution over possible next tokens with a softmax function, and sample one, repeating the process until a passage is complete. Sampling settings such as temperature control the trade-off between predictability and surprise, since low temperatures favor safe, common continuations while higher temperatures produce more unusual word choices at the cost of coherence. Because a model has absorbed patterns from vast amounts of fiction, screenplays, and conversation, it can imitate genres, voices, and story conventions with little instruction, and prompt engineering, few-shot examples, and fine-tuning can push it toward a particular style.

Commercial models are also shaped by reinforcement learning from human feedback and similar alignment methods, which reward responses that people rate as helpful and agreeable. One plausible contributor to a pattern documented in recent evaluations, in which machine-written stories resolve conflict too quickly and drift toward upbeat endings, is this preference for agreeable output, although the causes have not been firmly established. Writers who use these tools often counteract the tendency by explicitly requesting setbacks, moral ambiguity, or unresolved endings, or by using models tuned specifically for fiction.

Memory, Coherence, and World State

The hardest technical problem in machine storytelling is long-range coherence. A model sees only what fits in its context window, and even when that window holds hundreds of pages, models can lose track of who knows what, contradict earlier events, or forget a planted clue; the authors of DeepMind's Dramatron noted that when they wrote their paper, state-of-the-art context windows held at most about 2,048 tokens, roughly 1,500 words. Keeping characters straight also requires coreference resolution, the task of determining which name or pronoun refers to which entity across a long text. Story tools therefore add external memory: NovelAI's Lorebook stores background facts and inserts them when relevant, interactive fiction platforms track world state and prior choices, and research systems use retrieval-augmented generation or knowledge graphs to give the model only the facts a scene needs.

Persistent memory also underlies emergent narrative, in which stories arise from interacting characters rather than a pre-written plot. In the 2023 Generative Agents study by Stanford and Google researchers, 25 agents built on a language model stored their experiences in natural language, periodically synthesized them into higher-level reflections, and retrieved them to plan behavior in a sandbox town inspired by The Sims. Seeded with a single idea that one agent wanted to throw a Valentine's Day party, the agents spread invitations, made new acquaintances, asked each other on dates, and coordinated to arrive together, and ablation tests showed that observation, planning, and reflection each contributed to believability. The work won a best paper award at UIST 2023 and has become a reference design for AI agents in games and simulations, as well as a vivid example of emergence in narrative; lead author Joon Sung Park later grounded agents in interviews with more than 1,000 real people for social science research, reporting that they replicated participants' attitudes and behaviors 85 percent as well as the participants replicated their own.

Planning, Hierarchy, and Multi-Agent Writing

Because a single pass rarely produces a well-structured long story, many systems plan before they write. Dramatron, developed at DeepMind and evaluated in a CHI 2023 paper, starts from a one-sentence log line and uses prompt chaining to generate, in order, a title, characters, story beats, location descriptions, and dialogue, building structural context at each level before drafting scenes. In a study with 15 theater and film professionals, participants found it useful for world building, idea generation, and exploring alternative story paths, but also described outputs as formulaic and noted that its top-down structure did not match every writer's process. Scripts co-written with Dramatron were heavily edited by human writers and staged at the Edmonton International Fringe Theatre Festival in August 2022 in a production called Plays by Bots.

Agents' Room, presented by Google DeepMind researchers at ICLR 2025, extends the idea to a team of specialized agents modeled loosely on a television writers' room. A central orchestrator assigns subtasks to planning agents responsible for conflict, setting, plot, and character, and to writing agents that each draft one part of Freytag's pyramid (exposition, rising action, climax, falling action, and resolution), with all agents sharing a scratchpad. Using a new Tell Me A Story dataset of about 350 expert-written stories averaging roughly 2,000 words, the authors found that expert evaluators preferred Agents' Room stories over end-to-end baselines, although human-written stories were still preferred overall. Related research shows that making narrative structure explicit helps: a 2024 EMNLP study reported more than 40 percent improvement in diversity, suspense, and arousal when discourse features such as story arcs and turning points were integrated into generation, an approach that resembles chain-of-thought prompting applied to plot.

Narrative Structure and Emotional Arcs

AI research on storytelling draws heavily on narrative theory, from Aristotle's emphasis on plot to Gustav Freytag's five-part dramatic pyramid and Kurt Vonnegut's argument that stories have shapes that can be drawn on graph paper. In 2016, University of Vermont researchers tested Vonnegut's idea computationally by applying sentiment analysis to 1,327 works of fiction from Project Gutenberg and tracking how emotional tone rose and fell across each text. Using three independent methods, matrix decomposition with singular value decomposition, hierarchical clustering, and a self-organizing map neural network, they identified six core emotional arcs: rags to riches, riches to rags, man in a hole, Icarus, Cinderella, and Oedipus. They also found that particular arcs were associated with more downloads, suggesting that emotional trajectory, not just topic, affects how audiences respond.

The same lens is now used to diagnose weaknesses in machine-generated stories. In Are Large Language Models Capable of Generating Human-Level Narratives? (EMNLP 2024), researchers from UCLA and the University of Southern California analyzed story arcs, turning points, and the affective dimensions of arousal and valence in movie synopses and in stories written by models such as GPT-4 and Claude. Human-written stories were suspenseful, arousing, and diverse in structure, while the LLM stories were homogeneously positive, lacked tension, and struggled to develop major setbacks and climaxes; most models also fell short of humans at identifying these structures when reading. The authors argue that weak narrative comprehension may limit generation, and their gains from explicit structural guidance suggest one practical remedy.

Multimodal Storytelling: Images, Voice, and Video

Stories are increasingly told in pictures, sound, and motion as well as words. Image and video generators based on diffusion models learn to turn random noise into images or frames, usually working in a compressed latent space, while speech synthesis supplies narration and character voices and speech recognition turns spoken interviews into text that can be analyzed. The central storytelling challenge in these media is consistency: a character must look and sound the same from shot to shot, and lighting, setting, and props must persist across a scene. Recent tools address this with reference images, character libraries, and scene-building interfaces, but long-form visual continuity remains less reliable than human production.

Video models gained synchronized sound in 2025. Google introduced Veo 3 with native generation of dialogue, sound effects, and ambient audio at its I/O conference in May 2025, together with Flow, a filmmaking tool that lets users generate eight-second clips, connect them into longer sequences, and keep characters consistent from one clip to the next; by October, when it released Veo 3.1, Google reported that more than 275 million videos had been generated in Flow. OpenAI released Sora 2 on September 30, 2025, with synchronized audio and a consent-based cameo feature for inserting verified likenesses into scenes, alongside a social video app. That app proved short-lived: OpenAI announced on March 24, 2026 that it would discontinue Sora's consumer app and API, closing the app on April 26 and the API on September 24, 2026, after describing the service as resource-intensive with little revenue and amid declining downloads, and the decision also ended a content partnership with Disney.

Consumer storytelling features have spread into general assistants. In August 2025 Google added Storybook to the Gemini app, which turns a prompt, photos, or a child's drawing into a ten-page illustrated book with read-aloud narration in more than 45 languages, initially limited to users 18 and older. Google DeepMind's Genie 3, announced the same month, points further ahead: this world model generates navigable environments from a text prompt in real time at 720p and 24 frames per second, keeps them largely consistent for a few minutes, and supports promptable world events such as changing the weather or adding objects. DeepMind presents Genie 3 primarily as a research tool for training embodied agents, but its ability to generate explorable scenes on demand suggests future forms of interactive narrative that are rendered rather than authored in advance, even though interaction is currently limited to minutes rather than hours.

Applications

Storytelling shapes how AI is designed, explained, taught, marketed, and experienced, and it also describes a growing set of things AI does on its own. The applications below begin with the ways stories help people build and understand AI, then turn to the creative fields where AI increasingly participates in telling the story itself. Across all of them, the most successful deployments use AI to bridge the gap between technical systems and the people who use them, rather than to replace human judgment about what a story should mean.

Design and Product Development

Storytelling plays a central role in design, well beyond its traditional use in literature and entertainment. Designers use narrative to communicate the purpose and vision behind a product, to make complex ideas relatable, and to evoke the emotions that make an experience memorable. User personas and journey maps let teams imagine a product through a specific person's goals and frustrations, and agile teams capture requirements as user stories that describe who needs a feature and why. This empathetic, narrative approach helps designers build experiences that resonate with users and fosters a lasting relationship between people and the products they use.

AI now supports this work at several stages. Generative design tools produce options that satisfy stated criteria and constraints, letting designers explore many possibilities quickly, while image and pattern recognition help teams analyze user preferences and market trends. Analysis of user behavior enables personalization, virtual prototyping saves time and resources, and natural language processing makes more intuitive conversational interfaces possible. These capabilities support data-driven decisions and continuous improvement, but they work best when designers treat AI output as raw material for a human-authored story about the product rather than as the design itself.

Explainability and Interpretability

Explainability and interpretability are among the most practical uses of storytelling in AI. Models, especially deep learning neural networks, are often considered black boxes because their decision-making processes are not transparent. Presenting a decision as a story, with inputs, the factors that mattered, and the resulting output, helps users and stakeholders understand why a particular decision was made, which can increase appropriate trust and adoption.

Researchers are now using language models to write these narratives automatically. MIT's EXPLINGO system, presented at the 2024 IEEE International Conference on Big Data, takes SHAP explanations, which assign each input feature a value describing how much it pushed a prediction up or down, and converts them into short plain-language paragraphs. A NARRATOR component is prompted with three to five example narratives so it can mimic a desired style, and a GRADER component uses a language model to score each narrative for conciseness, accuracy, completeness, and fluency. Because the LLM only rephrases an existing explanation rather than inventing one, the design limits opportunities for error; even so, the researchers found that adding more style rules increased the risk of inaccuracies, a reminder that a fluent story can still misrepresent what a model actually did.

Data storytelling applies the same principle to analytics. Instead of presenting dashboards full of numbers, analysts structure findings as a narrative with context, a problem or anomaly, and a recommended action, using visualization to carry the evidence. Generative AI tools increasingly draft these summaries automatically, which saves time but raises the same accuracy concern as narrative explanations: readers tend to trust a coherent story, so the underlying numbers still need to be checked.

Human-AI Interaction (HAII)

As AI systems spread across industries, smooth human-AI interaction (HAII) has become a design priority. Chatbots and virtual assistants with conversational capabilities can use storytelling to make interactions more natural and relatable, for example by explaining a recommendation through a short scenario or by maintaining a consistent persona. AI-powered tools now generate many kinds of narrative content, including news articles, social media posts, and screenplays, and AI can also analyze and interpret stories to provide insights into audience engagement and preferences.

Researchers have studied how people and AI write together by designing large interaction datasets that log suggestions, acceptances, and edits in collaborative writing sessions, such as Stanford's CoAuthor dataset of writing sessions with GPT-3. These experiments aim to understand when AI support boosts creativity and productivity and when it takes over too much of the author's voice. Related work examines how human-AI interaction differs across languages and situations, and as AI becomes more interactive and embodied, researchers are asking how individual differences and different AI embodiments, from text chat to voiced avatars and social robots, shape how people perceive and engage with machine storytellers.

Education

Storytelling plays an important role in AI education and outreach. Complex AI concepts and algorithms can be daunting for beginners, and explaining AI principles, history, and real-world applications through stories makes the subject more accessible and engaging for learners of all ages. AI-powered interactive storytelling platforms can also serve as educational tools that create personalized learning experiences, such as stories that adapt vocabulary and pacing to a reader's level or let students explore historical events through characters' choices, an approach closely related to game-based learning.

General-purpose assistants have made personalized stories easy to produce for classrooms and families. Google's Gemini Storybook explicitly suggests educational uses, such as a story that explains the solar system to a five-year-old, and constructionist research presented through the Connected Learning Alliance has shown educators building storytelling projects with machine learning tools that turn everyday objects into interactive toys. Such projects can teach children how AI works by letting them build with it, although researchers also report troubleshooting challenges and resource needs when these activities move into classrooms.

Storytelling toys are a fast-growing and contested category. Mattel announced a strategic collaboration with OpenAI in June 2025 to develop AI-powered products and experiences based on its brands, and smaller companies began selling plush companions that converse with and tell stories to children. In late 2025 the U.S. PIRG Education Fund's annual Trouble in Toyland report and a follow-up study documented AI toys that directed children toward dangerous household items and, in one teddy bear's case, produced sexually explicit responses; the maker temporarily delisted the product, and OpenAI said it had suspended the developer for violating policies that prohibit endangering or sexualizing minors. These findings show why child-directed storytelling requires stronger safeguards than adult creative tools.

Marketing and Advertising

Storytelling has long been the cornerstone of effective marketing and advertising campaigns. With AI-powered personalization and recommendation systems, brands can tell individualized stories, analyzing user data to understand preferences and behavior and delivering tailored content and advertisements that resonate with each person. Generative tools now extend this to production itself: marketing platforms such as Jasper draft campaign copy in a brand's voice, avatar video services such as Synthesia turn scripts into presenter-led videos in many languages, and unscripted video tools turn customer interviews into testimonials.

The same capabilities raise questions about authenticity. Testimonials, case studies, and founder stories persuade because audiences assume they are real, so undisclosed synthetic voices, avatars, or composited interviews can undermine trust, and personalized narratives built from behavioral data raise privacy concerns. Many organizations therefore pair AI production with human review and clear labeling, using automation to scale the telling of real stories rather than to invent them.

Virtual and Augmented Reality

AI can be combined with virtual reality (VR) and augmented reality (AR) to create immersive and interactive storytelling experiences. Algorithms can adapt the narrative to a user's actions and preferences, providing a dynamic and personalized story inside a virtual environment, and generative models can populate scenes with dialogue, sound, and 3D models on demand. TouchStar, a research project described in an ACM publication, illustrates the approach: this interactive VR film about a visually impaired girl uses gesture and voice input and generates environments, sound effects, 3D models, and dialogue at key narrative points in response to the viewer. World models such as Genie 3 suggest that entire explorable settings may eventually be generated around a story as it unfolds, connecting AI storytelling to the broader metaverse.

Content Creation and Publishing

AI can aid in content creation by helping write stories, generate artwork, and compose music. Writing assistants range from tools that work like an advanced autocomplete, continuing a passage in the author's style, to systems that outline, draft, and revise entire chapters, while image and music generators supply illustrations and soundtracks. Used well, these tools expand creative possibilities for artists and content creators, helping them brainstorm, overcome blank-page paralysis, and prototype ideas quickly.

The ease of generating text has also strained publishing systems. In February 2023 the science fiction magazine Clarkesworld temporarily closed submissions after receiving about 500 machine-written stories alongside 700 legitimate ones, and in September 2023 Amazon's Kindle Direct Publishing began requiring authors to disclose AI-generated text, images, or translations and capped new title creation at three per day. Platforms built around reader data have embraced AI more fully: Inkitt, which selects popular self-published stories for its Galatea app, raised $37 million in 2024 with plans for AI-generated and personalized fiction, audiobooks, and video. Synthetic narration has likewise made audiobook production faster and cheaper, while raising concerns among professional narrators and voice actors.

Film and television are adopting generative tools more cautiously. Netflix co-CEO Ted Sarandos said in July 2025 that the Argentine series El Eternauta contained the company's first generative AI final footage, a building-collapse sequence completed about ten times faster than with traditional visual effects workflows. OpenAI backed Critterz, an animated feature made largely with its tools by Vertigo Films and Native Foreign on a budget of about $30 million and a nine-month schedule, with producers targeting a premiere at the Cannes Film Festival in May 2026. In June 2026 Google DeepMind formed a research partnership with the independent studio A24, accompanied by a reported $75 million investment, and A24 said the resulting tools would support filmmakers' workflows rather than generate films from prompts, a distinction that runs through debates about AI in books, radio, and movies.

Gaming

Storytelling has always been a vital part of video games, and AI can make game narratives dynamic and adaptive, responding to the player's choices and actions. Procedural systems can generate quests and storylines that increase replay value, while language-model-driven non-player characters (NPCs) can hold open-ended conversations instead of selecting lines from fixed dialogue trees, as discussed under generative AI in game development. Middleware from companies such as Inworld and NVIDIA's ACE tools connects game engines to language, speech, and facial animation models, and Ubisoft's research team moved from its NEO NPC prototypes, shown at GDC 2024, to Teammates, a playable first-person experiment revealed in November 2025 in which two squadmates and an assistant named Jaspar respond to the player's voice in real time using Google Gemini and internal middleware.

Narrative designers stress that generative characters still need authorship. Teammates' narrative director described an approach in which the AI characters can improvise while the team still sets the rules and directs the story, and Meaning Machine, a Bristol studio, calls its control technology Authored AI. A publicly funded University of Bristol study of the studio's murder-mystery game Dead Meat, presented at GDC 2026, reported that 95 percent of participants in the first 68 sessions found the experience enjoyable and 97 percent found it rewarding, while some players also found the freedom overwhelming; the results are preliminary, the full 122-session study was still to be published, and critics noted that the research was conducted in collaboration with the studio whose games were tested. The industry remains divided: in the 2026 State of the Game Industry survey of more than 2,300 professionals, 52 percent said generative AI was having a negative impact on the industry, including 63 percent of respondents working in game design and narrative.

Many of these ideas have roots in tabletop play. Storytelling games without a game master, such as Fiasco and Microscope, distribute authorship among players, and legacy games such as The King's Dilemma carry consequences from one session to the next. Studying these mechanics, as the AI creativity and game design case studies do, helps designers decide which narrative decisions to hand to an AI and which to keep in human hands.

Interactive Fiction, Role-Play, and AI Companions

Text-based interactive fiction was the first place many people experienced AI storytelling. AI Dungeon, created by Nick Walton in 2019 and later developed by his company Latitude, used GPT-2 to act as an open-ended game master that improvised a story around anything the player typed, and it went on to attract millions of players. In April 2026 Latitude launched Voyage, a platform that lets users design their own AI role-playing games with persistent characters and a logic-tracking World Engine, reflecting lessons about memory and rules learned from AI Dungeon. Similar dynamics drive role-play on Character.AI, companion apps such as Replika, and open-source front ends such as SillyTavern, where users co-write ongoing stories with persistent characters.

The intimacy of character-based storytelling raises distinctive risks, particularly for young users. Facing multiple wrongful-death lawsuits and regulatory scrutiny, Character.AI ended open-ended chat for users under 18 in November 2025 and introduced Stories, a guided format in which teens make choices within branching narratives instead of conversing freely with characters. The change reflects growing concern, discussed under AI companions and mental health, that immersive fiction can slide into emotionally dependent relationships, and that the line between storytelling and companionship has to be designed deliberately.

Research Findings and Limitations

Measuring Machine Creativity

Judging whether a machine-written story is good is itself a research problem. In Art or Artifice? Large Language Models and the False Promise of Creativity (CHI 2024), Columbia University and Salesforce researchers adapted the Torrance Tests of Creative Thinking into 14 binary Torrance Tests of Creative Writing organized around fluency, flexibility, originality, and elaboration. When ten creative-writing experts applied the tests to 48 stories, 12 by professionals and 36 by leading LLMs, the model-written stories passed three to ten times fewer tests, and when the researchers tried using LLMs as judges, none of the models' assessments positively correlated with the experts'. The second finding matters because many products and benchmarks rely on automated scoring by language models, a practice that may not capture literary quality.

Other studies reach more favorable conclusions under different conditions. A 2024 study measuring the psychological depth of stories, their capacity to evoke empathy, engagement, and emotional response, reported results far more favorable to LLMs and attributed the divergence from earlier work partly to much shorter stories (about 450 words rather than 1,400), since models struggle more with long texts. Public reactions are similarly split: when OpenAI chief executive Sam Altman shared a metafictional short story about AI and grief from an unreleased creative-writing model in March 2025, novelist Jeanette Winterson praised it in The Guardian, while other writers, including Dave Eggers, sharply criticized it. Taken together, the evidence suggests that current models can produce polished short passages but remain weaker at sustained structure, tension, and originality, and that results depend heavily on length, genre, prompting, and who is doing the evaluating.

Creativity Gains and Homogenization

One of the most cited experiments on AI-assisted storytelling is a 2024 Science Advances study by Anil Doshi of University College London and Oliver Hauser of the University of Exeter. In a preregistered online experiment, 293 writers composed eight-sentence stories with or without access to AI-generated story ideas, and 600 evaluators who did not know the conditions rated the results. Access to AI ideas made stories more creative, better written, and more enjoyable, with almost all of the gains going to the least creative writers; however, stories written with AI help were more similar to one another than stories written without it, according to computational measures of textual similarity. The authors describe a social dilemma in which individual writers are better off while the collective pool of novel content becomes narrower, a pattern that could compound as more creative work is drafted with the same few models.

Recent Developments

Several trends defined AI storytelling between 2025 and late 2026. The first is a shift from one-shot generation toward structured, authored systems: Agents' Room in research, Latitude's World Engine and Meaning Machine's Authored AI in games, and editing tools such as Quickture in professional production all constrain generative models with explicit plans, world state, or human-defined story goals. The second is commercial consolidation around workflows that pay. OpenAI shut down the Sora consumer app, ending a short-lived Disney character-licensing arrangement, while enterprise-focused companies grew; Synthesia raised $200 million at a $4 billion valuation in January 2026 to move from scripted avatar videos toward conversational, agent-driven learning experiences.

The third trend is a maturing body of evidence and rules. Peer-reviewed studies now document both the benefits of AI assistance and its costs to diversity and narrative tension, regulators and courts have begun to define the limits of copyright for AI-generated works and AI training, and Hollywood guilds negotiated new contracts in 2026 with expanded AI provisions. At the same time, professional sentiment has hardened, especially among artists and narrative designers, so the field's near-term direction will likely depend less on raw model capability than on whether tools can demonstrably support human storytellers rather than displace them.

Ethics, Law, and Labor

Copyright and Authorship

Copyright law affects AI storytelling at both ends of the pipeline: the works used to train models and the stories models produce. On the output side, the U.S. Copyright Office's January 29, 2025 report, Copyright and Artificial Intelligence, Part 2: Copyrightability, reaffirmed that copyright protects only human authorship, concluding that material generated by AI without sufficient human control over its expressive elements is not protected, that prompts alone are generally not enough, and that human selection, arrangement, and modification of AI output can be protected. Federal courts have taken the same position, upholding the Copyright Office's refusal to register a work listed as created autonomously by an AI system in Thaler v. Perlmutter. For storytellers, this means the human contributions to an AI-assisted work, such as plot choices, revisions, and arrangement, are what can be owned.

On the input side, the most consequential case so far is Bartz v. Anthropic, a class action by authors against Anthropic. In June 2025 Judge William Alsup ruled that training on lawfully purchased books was fair use but that downloading and keeping millions of pirated books in a central library was not, and the parties then reached a $1.5 billion settlement, roughly $3,000 per work for about 465,000 books, which the court granted final approval on July 20, 2026. Described as the largest copyright recovery on record, the settlement covers only past acquisition and copying of the books, leaving claims about model outputs and dozens of other AI copyright suits unresolved, so the legality of training on fiction without permission remains contested beyond the facts of that case. Related legal and policy questions are covered under risk, safety, compliance, and regulation and AI and the law.

Labor and Industry Response

Writers, performers, and game developers have pushed for protections as AI tools improve. After the 2023 Hollywood strikes established the first limits on studio use of AI, the Writers Guild of America reached a tentative four-year agreement with the studios in April 2026 that was expected to include new rules on AI, such as licensing for AI training, while actors' and directors' unions negotiated their own 2026 contracts. In games, the share of professionals who said generative AI was having a negative impact rose from 18 percent in 2024 to 30 percent in 2025 and 52 percent in 2026, even as about 36 percent reported using generative AI tools in their work. Unscripted television presents its own issues: when AI helps editors assemble interviews into story beats, productions must guard against misrepresenting what real people said, a long-standing ethical concern in reality editing that automation could amplify.

Safety, Consent, and Authenticity

Generative storytelling can also be misused. Realistic synthetic video and voices make it easier to fabricate events or put words in real people's mouths, which is why likeness controls such as consent-based cameos, provenance labels, and disclosure policies have become standard features of responsible tools, a connection explored further under deepfakes. Copyrighted characters appeared in user-generated Sora videos within days of the app's launch, prompting OpenAI to promise more granular controls for rights holders, and children's products have shown how quickly open-ended story generation can drift into unsafe territory. Responsible practice, discussed further under AI ethics, typically combines age-appropriate design, human review for public-facing content, clear labeling of synthetic media, and respect for the consent of anyone whose story, voice, or face is used.

Further Resources

Additional Viewing

Load video
YouTube

In this 2016 TED talk, Princeton neuroscientist Uri Hasson describes experiments showing that listeners' brain responses align with a storyteller's when a story is understood, and that the alignment weakens when it is not. Although shorter than the other selections, it provides the scientific background for the introduction's discussion of why narrative is such an effective vehicle for communication.

Load video
YouTube

In The Quest for Automated Story Generation, a 2021 Waterloo.AI seminar, Georgia Tech professor Mark Riedl makes the case for studying storytelling through AI and focuses on two problems that run through this article: steering neural text generation toward goals and maintaining long-term causal relationships across a story. The talk connects the symbolic planning tradition described in the history section with the neural methods behind current language models.

Load video
YouTube

Stanford researcher Joon Sung Park explains the generative agents architecture in this hour-long 2023 talk for the London Machine Learning Meetup. He walks through how agents record experiences, consolidate them into reflections, and retrieve them to plan, and how a single seeded intention produced a Valentine's Day party in the simulated town, illustrating the memory and emergent narrative mechanisms described under How AI Generates Stories.

Load video
YouTube

Dramatica's 2022 introduction to Subtxt demonstrates an AI-assisted tool that helps writers structure a story around dramatic principles, character connections, and the argument the story makes. It complements this article's discussion of narrative structure and story-planning tools.

Load video
YouTube

Piotr Mirowski, a DeepMind staff research scientist and founder of the improvisational theater company Improbotics, presents Live Theatre Performance with Artificial Intelligence at a November 2023 UCL Interaction Centre seminar. He discusses generative AI as a creativity support tool, Improbotics' use of AI in improvised comedy since 2016, and DeepMind's evaluation of tools for co-writing screenplays and plays, making the talk a companion to the Dramatron discussion and the unscripted storytelling section.

Load video
YouTube

This Connected Learning Alliance presentation describes research in which educators created storytelling projects with a machine learning tool that turns everyday objects into interactive toys. It supports the Education section's discussion of constructionist approaches to teaching AI through stories.

Load video
YouTube

In this 2021 episode of EmergenTeck's Tech Storytelling series, Tom Allen, founder of The AI Journal, discusses AI in sales and marketing. It relates to the Marketing and Advertising section's discussion of AI-supported brand storytelling.

Load video
YouTube

In this 2017 GDC talk, NHTV Breda professor Mata Haggis offers practical techniques for building drama in indie games and shows how narrative craft strengthens overall game design. It provides the human narrative-design foundation that AI-driven game storytelling builds on.

Load video
YouTube

This GDC 2026 session, published by NVIDIA Game Developer, presents the University of Bristol and Meaning Machine study of AI-driven NPCs, including the design of authored AI systems and their integration with NVIDIA's In-Game Inferencing SDK. It is useful alongside the Gaming section for weighing early player-response evidence against the industry skepticism reflected in the GDC developer survey.

Unscripted Storytelling

Unscripted storytelling does not follow a predefined script; it relies on the spontaneity and creativity of storytellers and audiences, and it appears in reality television, documentaries, podcasts, improv comedy, and personal oral histories. Its raw material is real speech and behavior, which makes it authentic but also expensive to shape, because productions routinely record far more footage than anyone has time to watch and story producers must find the arc hidden inside hours of interviews. AI can generate text, images, audio, and video from input data, and the emerging field of AI-assisted unscripted storytelling applies those capabilities to finding, shaping, and delivering real people's stories. For example, AI can:

  • Conduct interviews with real or virtual characters and extract relevant information and emotions from their responses.
  • Craft stories based on the interview data and other sources, such as images, music, and graphics.
  • Edit and score the stories using techniques such as natural language processing, computer vision, and machine learning.
  • Present the stories to the audience in an engaging and interactive way, such as through a web platform, a mobile app, or a virtual reality headset.

Live performance offers a parallel tradition. The theater company Improbotics, founded by Piotr Mirowski, has used AI in improvised comedy since 2016, with human performers improvising alongside machine-generated dialogue, and party games such as Pitchstorm and Snake Oil show how much everyday storytelling consists of improvised pitching and persuasion. In both settings the most interesting work happens when humans and machines react to each other, rather than when either side simply executes a plan.

Orson

Story As A Service (The new SAAS) - Orson

Orson is a story-as-a-service startup that uses Generative AI and the science of video storytelling to help people share their stories and build meaningful human connections. It was founded in 2020 by John Ehrhard, co-owner and executive producer of the unscripted television company OAKZ Media, and was based in Winter Garden, Florida. In June 2023 the company announced a $3 million round led by Cyan Banister and Long Journey Ventures, bringing its total funding to $8 million, to scale the science of unscripted storytelling, expand production of its StoryStrip technology, and launch new products. Orson's central idea was to translate decades of unscripted television directing experience into software, so that a person could tell a compelling story simply by answering a director's questions.

Orson positioned itself as specialized for story-driven content rather than general-purpose video generation, emphasizing personalization around each subject's own experiences rather than generic templates. Its consumer offerings included StoryShop.ai and the Love Story Shop for engagement and wedding videos, and it formed partnerships to embed its directing engine in other companies' engagement platforms, including an alliance with Sirqul. Early StoryShop packages still used prerecorded questions from a human director, because, as Ehrhard told Fast Company, autogenerated avatar interviewers could seem stilted, and the company was building its own datasets of sample interviews for AI to learn from. When checked in October 2026, the heyorson.com address redirected to the site of XQai, an "execution intelligence" company whose Spotlight product offers AI-produced customer testimonials, HR screening videos, and employee features; the sources reviewed for this article do not document the relationship between the two companies.

Orson StoryShop automates video creation with the help of virtual directors. StoryShop applies lessons from producing unscripted television to an automated director that conducts interviews, crafts stories, and edits them together with music, imagery, and graphics into episodes that help individuals tell their stories. Ehrhard described the goal as lowering the barrier not just to video editing, which many tools already simplify, but to unscripted video storytelling itself.

Orson StoryStrips™ are short, unscripted video stories created by interviewing real people and using AI to edit the interviews into finished videos with graphics, music, and branding. Each StoryStrip tells a complete story while also fitting into a larger narrative, so a series of strips can be released episodically, much like the comic strips the product's name evokes. According to the company, its AI-enabled directors and StoryStrip technology made the process fully automated and scalable, allowing story-driven video to be produced from interviews at volumes that traditional crews could not match.

The Orson AI Director is the system that guides the storytelling process. It conducts interviews, reads facial cues, and asks follow-up questions based on the subject's answers, then uses that material to shape a story, select supporting visuals, and assemble an edit that is the right length and flows smoothly. One of the capabilities Ehrhard has highlighted is the AI's ability to recognize the stopping point, the moment when an interview has achieved its emotional goals according to the guidelines the system was trained on. Like any automated director, it can misread instructions, choose imperfect footage, or make edits a human would not, so human review remains part of the workflow. Its core capabilities can be summarized as follows:

The AI Director can also adapt to different genres, formats, and styles of storytelling based on the user's preferences and feedback. The process ends when the user is satisfied with the final video or when the system reaches a predefined time limit or quality threshold, which echoes Ehrhard's emphasis on teaching the AI to recognize when a story's goals have been met. Defining that stopping rule is a design decision with creative consequences, because stopping too early leaves a story thin while continuing too long can make a subject feel interrogated rather than heard.

Orson's automation selects music and visuals that sync with the story at the right moments. Public descriptions of the platform are high level, but systems of this kind typically combine signals from the interview itself with learned patterns and human judgment. The main techniques are:

  • Audio analysis: analyzing the audio of the interviews to identify the key moments in the story, including changes in pitch, volume, and rhythm.
  • Visual analysis: analyzing the visuals of the interviews to identify key moments, including changes in facial expressions, body language, and scenery.
  • Machine Learning (ML): learning the patterns of how music and visuals are typically used to tell stories, so that the system can select music and visuals likely to be effective in a particular story.
  • Human curation: human experts review the selections made by the automation and make adjustments as needed.

Orson's AI Director relies on Natural Language Processing (NLP), the branch of AI that helps computers understand, interpret, and generate human language, to analyze and produce the text and speech behind its video stories. NLP draws on computer science and computational linguistics, and an interview-to-story pipeline typically chains several NLP tasks together. Because errors in early steps propagate downstream, the quality of the final story depends on every link in that chain, which typically includes the following tasks:

  • Speech recognition: reliably converting voice data into text, which is required for any system that conducts spoken interviews or answers spoken questions.
  • Part-of-speech tagging: determining the grammatical role of a word from its use and context, for example identifying make as a verb in "I can make a paper plane" and as a noun in "What make of car do you own?" Tagging supports downstream tasks such as extracting who did what in a subject's account.
  • Word sense disambiguation: selecting the intended meaning of a word with multiple meanings through semantic analysis of its context, which helps a system interpret what a speaker actually meant.
  • Coreference resolution: linking pronouns and names to the people and things they refer to, so that the story keeps track of who is who across a long interview.
  • Sentiment analysis and summarization: identifying emotional high points and condensing hours of material into the moments that carry the story.

Feedback loops are central to how a system like Orson improves over time. Each finished video generates signals about what worked, and those signals can be routed back into the system at several levels. The main feedback channels are:

  • User feedback: collecting feedback from users about the videos the system generates, which is used to improve the selection of music and visuals, the pacing of the video, and its overall emotional impact.
  • Data feedback: collecting data about how users interact with the videos, which is used to improve the algorithms that select music and visuals and set pacing.
  • Human curation: expert review of the automation's selections, which also provides feedback that can improve the automation in the future, a form of human-in-the-loop learning.

Orson-style unscripted video can serve many purposes beyond entertainment. The same interview-driven process that produces a personal documentary can produce a brand story, a lesson, or an internal announcement, with the AI Director adjusting questions and editing style to the goal. Common use cases include:

  • Creating marketing and advertising content: engaging, personalized marketing and advertising content that resonates with target audiences, for example video ads that tell the stories of a company's customers or employees.
  • Creating educational content: content that is both engaging and informative, for example videos in which a school explains complex concepts in a way students can understand.
  • Creating personal content: videos that document life events, preserve family history, or tell personal stories, such as a wedding day or a journey through cancer treatment.
  • Creating customer testimonials: testimonials that are more engaging and personal than traditional formats, in which customers tell their own stories about how a company's products or services helped them.
  • Creating employee onboarding content: human resources videos that introduce new employees to the company's culture and values in a more engaging way than traditional onboarding materials.
  • Creating internal communications content: videos that communicate new policies or announce upcoming events more engagingly than email or slide presentations, supporting leadership communication.


Have some specific goals that were trying to achieve and then teach the AI on what those goals look like and how to recognize when those have been achieved; it's phenomenal. I think one of the most exciting moments I've had ... was seeing the AI recognize the stop point. I know that sounds weird, but that actually was one of the most impressive parts where it sees the goals achieved emotionally, right? It asked enough follow-up questions based on the guidelines and how we trained it or it thought the goal was achieved - John Ehrhard | Orson

Quickture

  • Quickture ... AI editing for unscripted, documentary, news, and sports production.

Quickture is an AI editing tool built specifically for unscripted production. It was launched in 2024 by Irad Eyal, a reality television producer with 25 years of experience whose credits include Southern Charm and Netflix's Floor Is Lava, and it works inside professional editing software to transcribe footage, label story beats, and assemble rough cuts from natural-language prompts. Its premise is that reality, documentary, news, and sports productions shoot far more material than editors can review, so finding the story is the bottleneck, and Eyal has demonstrated a heat map that scores every beat in the footage. The company raised $4.5 million in seed funding in April 2026, four months after a pre-seed round, and Emmy-winning editor James Carroll has described using it to work as a one-person story team on a docuseries with more than 150 characters and interviews of three to four hours each.

Toolbaz Free AI Story Generator

Toolbaz offers a free AI story generator as part of a larger suite of free AI writing tools. Users enter specific details and parameters, such as genre, theme, setting, characters, and plot points, and can adjust story length and creativity level, and the tool generates an original story based on them. Like most general-purpose generators, it is best suited to brainstorming, casual fiction, and role-playing prompts rather than polished long-form work.

NovelAI

  • NovelAI ... Driven by AI, painlessly construct unique stories, thrilling tales, seductive romances, or just fool around. Anything goes!

NovelAI, developed by the Delaware-based studio Anlatan and launched in June 2021, is a monthly subscription service for AI-assisted authorship, storytelling, and role-play that also includes anime-style image generation. Its text models are tuned specifically for fiction: Erato, released on September 23, 2024, is based on Meta's Llama 3 70B model with continued training on hundreds of billions of tokens from NovelAI's Nerdstash pretraining dataset, and a model based on GLM-4.6 followed in late 2025. Features such as the Lorebook, which stores background facts about characters and settings and injects them when relevant, address the long-story consistency problems described earlier, and the service emphasizes encrypted, private stories and fewer content restrictions than general-purpose assistants.

AI Dungeon

  • AI Dungeon ... A text-based adventure-story game you direct (and star in) while the AI brings it to life.

AI Dungeon is a text-based adventure game in which players choose a character, setting, and premise and the AI generates the story as they play, functioning as an endlessly improvising game master. Created by Nick Walton in 2019 using GPT-2 and developed by Latitude, it has run on a succession of models, from OpenAI's GPT series and AI21's Jurassic models to open models such as Mistral AI's Mixtral, a mixture-of-experts model, along with Latitude's own fine-tunes. Its history includes well-publicized controversies over inappropriate generated content, and in 2026 Latitude built on its experience with Voyage, a platform for user-created AI role-playing games that shares memberships with AI Dungeon.

Jasper

Jasper is an AI content platform from Jasper AI, Inc., which was founded in January 2021 in Austin, Texas, by Dave Rogenmoser, Chris Hull, and John Phillip Morgan; it is not a Google product. Originally launched under the names Conversion.ai and Jarvis, it became one of the first breakout generative AI companies, raising a $125 million Series A at a $1.5 billion valuation in 2022, and it has since repositioned itself as an enterprise marketing platform with brand-voice controls and specialized marketing agents. Its story generator remains one of many tools for producing short stories, scripts, and other creative text, but the company's main focus today is marketing content.

Inkitt

Inkitt, founded in Berlin in 2013 and now based in San Francisco, is a reader-powered publishing platform where authors upload stories for free and the company uses engagement data and proprietary algorithms to identify works with commercial potential. Selected stories are edited and sold through its Galatea app, mostly in mass-market genres such as romance, fantasy, crime, thrillers, and science fiction. In February 2024 Inkitt raised a $37 million Series C led by Khosla Ventures, bringing total funding to about $117 million, with plans to use large language models for AI-generated and personalized fiction built from users' ideas, machine translation, audiobooks, video, and games. That strategy has drawn criticism from writers who argue that leaning on generative AI for ghostwriting, editing, narration, and cover design devalues human authorship, making Inkitt a prominent test of data-driven, AI-assisted publishing.

StoryAi

  • StoryAi ... Interactive, choose-your-path story adventures with AI-generated video scenes.

StoryAi began as a web tool for generating stories from a description of what the story should be about. When reviewed in October 2026, the site presented StoryAI Adventures, an interactive choose-your-path game in which each decision triggers a short AI-generated video scene, with starting worlds in dark fantasy, science fiction horror, whimsical anime, and cyberpunk settings. Free play covers the opening rounds of each adventure, while a paid tier unlocks finales, generation of new story branches, and custom adventures, illustrating how simple text generators are evolving into multimodal interactive fiction.

Similar

Many AI video tools overlap with Orson's goal of turning text or real people's words into story-driven video, although most start from a script, article, or existing footage rather than from an AI-directed interview. The comparisons below focus on how each differs from that interview-driven approach, based on publicly available information as of 2026. Because features and pricing in this market change quickly, readers should confirm current capabilities on each vendor's site.

  • Synthesia: an AI video platform, founded in 2017 and based in London, that turns typed or pasted scripts into videos presented by digital avatars, with facial animation, synthetic voices, and translation into many languages; it is widely used for corporate training and communications. In January 2026 it raised $200 million in a Series E round led by GV at a $4 billion valuation and announced plans to extend its avatars from script readers into interactive agents. Synthesia is similar to Orson in that it creates story-driven content from text input, but by default it does not use real people's footage or direct a subject through interview questions.
  • Pictory: an AI video generator that turns articles, blog posts, scripts, URLs, or audio into videos for social media or websites, automatically selecting stock footage and music and adding captions, AI voiceovers, and optional AI avatars. Pictory is similar to Orson in that it creates engaging, informative content from text input, but it focuses on repurposing existing written or recorded material rather than on stories drawn from interviews.
  • Fliki: a text-to-video and text-to-speech platform that converts ideas, scripts, articles, or URLs into narrated videos, offering more than 2,500 AI voices, one-click translation into 80+ languages, and voice cloning. Fliki is similar to Orson in using AI to create creative, personalized content, but it centers on voiceover-driven video for social media, education, and marketing rather than on interviews with real people.
  • Descript: an AI video and podcast editor that transcribes recordings so users can edit audio and video by editing the text, with tools for captions, music, sound effects, and regenerated speech. Its Underlord agent, released in open beta in July 2025, carries out editing tasks from natural-language instructions and can generate video with AI models, so Descript now generates as well as edits content. Descript is closest to Orson's editing stage, but it leaves interviewing and story direction to the user.
  • Peech: an AI video platform for marketing teams, founded in Israel in 2020, that automatically edits and brands talking-head footage, webinars, podcasts, and testimonials and repurposes them into short clips at scale. Peech is similar to Orson in producing polished video from people speaking on camera, but it starts from footage a company already has rather than conducting interviews, and it focuses on content marketing rather than personal storytelling.
  • Visla: an AI video generator that turns scripts, text, documents, webpages, audio, images, and clips into videos, automatically choosing stock footage, music, voiceover, and subtitles; users can keep a script unchanged or have the AI rewrite it, and an AI Director Mode helps plan scenes as storyboards. Visla is similar to Orson in producing narrative-style video from text and media, but it does not use AI-led interviews to gather the story.


Additional Viewing

Load video
YouTube

A Conversation with Ron Howard | Runway AI Festival 2026 | Runway | June 2026 | duration not confirmed Oscar-winning director Ron Howard sits down with Runway co-founder and co-CEO Cristóbal Valenzuela for the fireside chat that opened Runway's fourth annual AI Festival at Alice Tully Hall in New York's Lincoln Center on June 11, 2026. Press coverage of the event reported Howard arguing that audiences, not studios or tool makers, will ultimately decide whether AI-made films succeed, while also describing the tools as democratizing filmmaking. He told the crowd that story is always king and predicted that AI tools will yield "a new aesthetic" rather than cheaper copies of existing films. The conversation is worth watching because it shows a mainstream Hollywood storyteller judging AI video tools by what audiences care about emotionally, not by visual spectacle alone.

Load video
YouTube

A Conversation with Roger Avary & Gala Avary | Runway AI Festival 2026 | Runway | June 2026 | duration not confirmed This conversation was recorded at the Los Angeles edition of Runway's AI Festival on June 18, 2026. It features Oscar-winning Pulp Fiction co-writer Roger Avary and his daughter Gala Avary, who also appears on the Video Archives Podcast he hosts with Quentin Tarantino. Roger Avary served as a 2026 festival juror and is CEO of General Cinema Dynamics, a technology company he says has three AI-driven feature films in active production. Earlier in 2026 he said publicly that adding AI to his projects unlocked financing he had struggled to find through traditional pitching, which made him one of the most visible and debated figures in AI filmmaking. The video helps explain how veteran screenwriters are adapting to generative tools and what that shift could mean for who gets to tell stories on screen.

Load video
YouTube

LLM Creative Writing: Any good in 2026? Tested: Opus 4.8, Fable 5, GPT 5.5 | Edward Thomson | June 2026 | duration not confirmed Independent developer Edward Thomson reruns a fiction benchmark he built in 2025. It gives each model named in the title, along with Gemini 3 Pro and Flash, the same story prompt ten times. His tool measures how much the resulting stories overlap in wording, meaning, and even character names, so a vague complaint about samey AI prose becomes something that can be counted. The project grew out of his work on long-form fiction tools such as NovelWriter and StoryDaemon, where repeated names and recycled metaphors become real problems over hundreds of pages. Because the code and full report are published on GitHub, viewers can check or extend the method themselves. It is a practical, data-driven look at one of the central limits of AI story generation: models tend to converge on the same ideas.

Load video
YouTube

The Alchemy of AI Storytelling | The Telly Awards | May 2026 | duration not confirmed This session is part of the Telly Awards' Throughlines event series for its 47th annual competition. It features Ari Kuschnir, founder of the production company m ss ng p eces and creator of widely shared AI short films. Kuschnir walks through his creative process and explains how AI tools can serve an artist's vision, with attention to cultural impact and social change. Organizers of his screenings say his AI experiments have drawn more than 100 million views, and his videos use generative imagery to picture alternative social and political futures. The talk shows how a working commercial director uses generative video for social commentary, which makes it a useful case study of AI storytelling beyond tool demos.

Load video
YouTube

Storycaster: An AI System for Immersive Room-based Storytelling (CHI 2026) | Microsoft Research | April 2026 | duration not confirmed This Microsoft Research video presents Storycaster, a generative AI system from a paper given as an oral presentation at CHI 2026, the leading conference on human-computer interaction. The system uses live camera feeds to project cylindrical imagery onto the walls of a physical room, so the story world blends with the real space instead of hiding it behind a headset. A narrator agent guides participants as they co-create a story by voice, and each scene comes with generated ambient audio, dialogue, and imagery. The project points toward shared forms of AI storytelling in which people shape a narrative together in the same room. It is a good example of research taking AI narrative out of the chat window and into physical, social spaces.

Load video
YouTube

The AI Behind Teammates, Ubisoft's Experimental Interactive Characters | NVIDIA Game Developer | April 2026 | duration not confirmed This session covers Teammates, Ubisoft's first playable generative AI research project, in which players lead a squad of proactive non-player characters with real-time voice commands. The developers explain how they chose and tuned AI models to make the characters more immersive, and give a technical overview of deploying the system for fast, real-time inference. Ubisoft has said its narrative team still writes the story and the character personalities, setting boundaries within which the AI characters improvise while staying true to the lore. The talk shows how a major studio tries to balance authored narrative with emergent, player-driven dialogue, one of the hardest problems in game storytelling.

Load video
YouTube

AI NPCs in Games | University Study, Player Behavior & NVIDIA In-Game Inferencing SDK Integration | NVIDIA Game Developer | April 2026 | duration not confirmed Researchers from the University of Bristol and the studio Meaning Machine present findings from a study of more than 100 players interacting with AI-driven non-player characters, in a session titled "What Good Are AI NPCs? Lessons from a Large-Scale Player Study". The speakers are Thomas Keane, Ben Ackland, Dr. Richard Cole, and Dr. Chris Bevan, and they examine how people actually behave when game characters can answer open-ended conversation. The session also covers how the team integrated NVIDIA's In-Game Inferencing SDK to run these characters. It matters because it replaces hype about conversational characters with evidence about what players actually do with them, which is essential for designing stories around AI characters.

Load video
YouTube

We Are The Art | Brandon Sanderson's Keynote Speech | Brandon Sanderson | January 2026 | duration not confirmed Fantasy author Brandon Sanderson, known for the Mistborn and Stormlight Archive series, uses this keynote to explain why he believes AI cannot replace human artists and authors. He takes AI image generators and large language models seriously, granting some points in their favor, before arguing that the creative process itself is what gives art its value. He draws on the motto "journey before destination" from his own books to argue that the growth a writer goes through while making a story matters as much as the finished product. The speech has been widely shared in online writing and fan communities as a statement of the human-centered case against generative fiction. It is a useful counterweight to tool-focused videos because it asks what storytelling is for in the first place.

Load video
YouTube

What is Voyage? Exploring Worlds Outside the Dungeon | AI Dungeon / Voyage | January 2026 | duration not confirmed Latitude, the studio behind the pioneering AI text adventure AI Dungeon, uses this livestream to introduce Voyage, its new platform for AI-driven worlds, during its open beta. The hosts discuss emergent storytelling and character dynamics, explain how Voyage differs from AI Dungeon, and walk through its scripting and creation tools for world builders. They also cover genre flexibility, the choice to put storytelling ahead of visuals, the launch timeline, and how community feedback shapes development. The stream's chapters run past the 44 minute mark, making it one of the more detailed first-party looks at how an AI storytelling company is evolving its design. It is valuable for anyone studying how AI game narrative is moving from pure text toward persistent, explorable worlds.

Load video
YouTube

The future of Novels and Generative AI | Cambridge University | December 2025 | duration not confirmed Dr Clementine Collett of Cambridge's Minderoo Centre for Technology and Democracy discusses her major survey of the UK fiction publishing industry with Ella, an English undergraduate at Murray Edwards College. The research behind the video gathered responses from 258 published novelists and 74 industry professionals. It found that 51 percent of the novelists think AI is likely to replace their work entirely, and 39 percent say their income has already suffered. Collett explains concerns about copyright, lost income, and the future of human-written novels, while noting that the findings were not entirely opposed to AI. She also discusses what the results mean for Britain's culture, economy, and creativity. The video gives a grounded, evidence-based picture of how the people who write and publish stories are actually experiencing generative AI.

References

Retrieved from "https://primo.ai/w/index.php?title=Storytelling&oldid=53421"