Scriptr is an AI collaborator trained on screenplays and movies to help creatives quickly ideate unique stories that have never been told before. It functions as a storytelling companion to assist with the creative process by generating contextually aware text and images. It intends to be integrated with industry-standard applications like Final Draft as the first step before narrowing down on details. This system leverages the power of natural language processing models to write content from user-written nodes that connect to create an infinite set of non-linear storylines. This project was created at Art Center College of Design for AI & Agents by Davis Brown and Greg Palkovics.
The editor view is where the user defines nodes that are bridged together with AI. In this view you can see how the nodes connect and the color gradient that forms between nodes of different styles. At the center of each line is an interpolation point. When the user clicks this point they can modify the story arch and iterate new ideas for their story. The user has the option of creating an extremely detailed node map or one that is more broad that lets the AI system iterate more variety.
Iteration is a key principle behind Scriptr. The iterate button is highlighted in blue along with the interpolation points to indicate that the AI can be used for idea generation. When the user clicks the iterate button they are provided with storyline alternatives to chose from. There is an infinite set of possibilities that can emerge through this process and whenever the user likes an idea they can easily save it or lock it from being modified from the next iteration.
Story arc intensity adjustment is another key feature inside of Scriptr. From our research we found that having this fine tune control to adjust the low and high points in a story was crucial for creatives to adjust their narrative. We looked at how different story arcs are used across movie and story genres to explore how this system could be used by our users.
The board view inside of Scriptr complements the node editor by generating visuals through machine learning based on keyword and pattern recognition from the storyline. The board view will create an abstract representation of each key event in the user's story along with visuals of characters, clothing, environments and miscellaneous objects. When an image is selected on the board it can be modified further to refine down to a specific style and appearance. Changes made in either the board or the editor will reflect between each other.
When the user exports their story created in Scriptr they are provided with multiple options on sharing their work. The primary export is as a PDF with text and image generated from the board and editor. The next form of export is as a video with narration performed by a natural voice AI. This video will transition between image frames off the storyboard with the text captioned below. The third export method is to open into Final Draft for secondary review and refinement. Images in the visual have been generated with ArtBreeder and the voice over is from LOVO.
This is our system diagram which outlines the process that the user interacts with the front-end interface. The AI trained inside Scriptr acquires new vocabulary through words, phrases, sentences and images that the user inputs. With this new vocabulary it is able to generate new content from patterns in the vocabulary as either a Text or Image story format.