Man and Machine:
AI as co-creator in storytelling

AI as Co-Creator in Storytelling: Character Awareness, Style & Script Structure in Generative AI

After more than half a century of research, artificial intelligence has finally entered its useful age. Creative industries were considered "untouchable" for a long time due to the human nature of creativity. However, with recent breakthroughs, AI is making great strides. Whether dealing with images, video, music, speech or text, deep learning has become the go-to solution for automation in almost any domain.

While AI can play the role of the artist itself, the most promising examples are those where human and machine collaborate — with AI filling the role of a co-creator.

State-of-the-art: today's script generators

The engine behind any text-generating AI is a language model — a statistical model that predicts and writes text by learning from large volumes of examples. In 2016, Sunspring was released, the first short film ever to be entirely written by an AI. While each sentence is grammatically correct, the dialogue reads as though vital context is missing — semantically almost devoid of meaning.

The common theme across AI-generated scripts is this: while language models have learned to write syntactically correct English, they seem to have little understanding of how a story is actually told. Characters lack consistent personalities. Settings are established then ignored. And stories fall apart at scale because the models cannot maintain coherence across scenes.

Beyond the state-of-the-art: introducing "story-awareness"

At ScriptBook, we put the storyline central. We spent years building deep expertise in AI that can analyze and comprehend screenplays. This strength allows us to transfer key elements from our decision support system to a generative AI — developing an engine that is genuinely more capable of generating stories. We call this concept "story-awareness," built across three dimensions.

Character awareness

A good script generator should understand that characters are defined not just by their names, but by their traits and personalities. A character's gender and looks don't change throughout a story, and shifting personalities should be gradual and sensible.

At ScriptBook, our algorithms distinguish different characters from each other, extract their traits, likeability, and emotions — building what we call a "DNA profile" of each character. In the script generator, we reverse this process: every time a character acts or speaks, the model is made aware of that character's profile. The result is more consistent, believable behaviour.

Style and theme awareness

A good story has a setting, theme, and genre — and doesn't suddenly shift halfway through. At ScriptBook, we make use of thousands of scripts spanning all conceivable genres. We extract a script's "fingerprint" containing story-level parameters and make the generator aware of the preferred fingerprint — not unlike requesting a preferred setting, theme, or genre from a human writer.

Script structure awareness

Finally, we incorporate a screenplay's structural elements — scene headers, dialogue formatting, location descriptions — explicitly into the language model. When we generate a scene set in a specific location, the generator is made aware of that context, and it is reflected consistently in the output.

Moving forward

At ScriptBook, we understand stories. Our generative engine plugs parameters from our existing script analysis AI into a language model, resulting in improved writing capabilities and story consistency. That said, our model is still far from becoming the next generation screenwriter.

The most promising examples in AI are those where humans and machines collaborate. Imagine an application in which the writer specifies a story's outline, and uses AI to write out the various scenes — an invaluable source of inspiration, if not a finished product.

And while we don't think such machines will replace human writers anytime soon, we are convinced it's only a matter of time before AI finds its way into the screenwriter's toolbox.

What this means in 2026

When this research was published, GPT-2 had just been released and "generative AI" was not yet a household term. The story-awareness concepts we developed at ScriptBook — character DNA profiling, style biasing, structural annotation — have since become standard approaches in large language models, though most implementations still struggle with the long-form consistency problems we identified.

ChatGPT, Claude, and Gemini can now generate coherent dialogue and scene descriptions. But ask them to maintain a character's voice consistently across a 90-page screenplay, or to honour the tone fingerprint of a specific director's style — and the limitations we described in 2019 are still very much present. The problem of story-awareness in AI remains unsolved at scale.

This research directly led to US Patent US20200334336A1 for the generation of scripted narratives — filed in 2019 and cited by Oracle, Tencent, and others.