Authors:
(1) Prerak Gandhi, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, prerakgandhi@cse.iitb.ac.in, and these authors contributed equally to this work;
(2) Vishal Pramanik, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, vishalpramanik,pb@cse.iitb.ac.in, and these authors contributed equally to this work;
(3) Pushpak Bhattacharyya, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai. Table of Links Abstract and Intro
Motivation
Related Work
Dataset
Experiments and Evaluation
Results and Analysis
Conclusion and Future Work
Limitations and References
A. Appendix Abstract Storytelling is the lifeline of the entertainment industry- movies, TV shows, and stand-up comedies, all need stories. A good and gripping script is the lifeline of storytelling and demands creativity and resource investment. Good scriptwriters are rare to find and often work under severe time pressure. Consequently, entertainment media are actively looking for automation. In this paper, we present an AIbased script-writing workbench called KUROSAWA which addresses the tasks of plot generation and script generation. Plot generation aims to generate a coherent and creative plot (600–800 words) given a prompt (15–40 words). Script generation, on the other hand, generates a scene (200–500 words) in a screenplay format from a brief description (15–40 words). Kurosawa needs data to train. We use a 4-act structure of storytelling to annotate the plot dataset manually. We create a dataset of 1000 manually annotated plots and their corresponding prompts/storylines and a gold-standard dataset of 1000 scenes with four main elements — scene headings, action lines, dialogues, and character names — tagged individually. We fine-tune GPT-3 with the above datasets to generate plots and scenes. These plots and scenes are first evaluated and then used by the scriptwriters of a large and famous media platform ErosNow[1]. We release the annotated datasets and the models trained on these datasets as a working benchmark for automatic movie plot and script generation. 1. Introduction Movies are one of the most popular sources of entertainment for people worldwide and can be a strong medium for education and social awareness. The impact and influence of film industries can be gauged from the fact that Hollywood movies invest *These authors contributed equally to this work 1 https://erosnow.com/ 100s of millions of dollars and often make boxoffice collections of billions of dollars. The first motion picture The Great Train Robbery, 1903— black & white with no sound— was created at the beginning of the 20th century. Since then, the art has gone through several transformations, and now people can instantly access 4K HD movies of their liking on any smart device. Throughout the history of film, two of the contributors to a film’s blockbuster success have been the quality of its plot and the manner of storytelling. The appeal of the movie decreases drastically if the viewers find the plot drably predictable. Writing a creative and exciting script is, therefore, a critical necessity and is extremely challenging. Add to this the constraints of time and budget, and the need for (at least partial) automation in script writing becomes obvious. AI-based story generation has been used before. Based on the engagement-reflection cognitive explanation of writing, the computer model MEXICA (Pérez and Sharples, 2001) generates frameworks for short tales. BRUTUS (Bringsjord and Ferrucci, 1999) creates short stories with predetermined themes like treachery. With the arrival of pre-trained transformer models, automatic story generation has got a shot in the arm. Transformer models like GPT-2 and GPT-3 are extensively used for text generation. These models have shown the capability of generating creative text, albeit sometimes with hallucinations (Zhao et al., 2020). Text generated by these models also sometimes lacks coherence and cohesiveness. On the other hand, template-based models can generate coherent text but lack creativity in generating new characters and events in the plot (Kale and Rastogi, 2020). The process of creating a movie generally starts with an idea which is then used to create a plot which is used as the base to build the movie script (Figure 1). Novel datasets are an important feature of this paper. We closely studied the plots and prompts of movies from Bollywood and Hollywood. Such plots and prompts were scraped from Wikipedia[2] and IMDb[3], respectively. The plots are then annotated using the 4-act story structure- an extension of the well-known 3-act structure (Field, 1979). The 4-act structure and the annotation methods are explained in detail in appendix A.5 and section 4, respectively. We introduce a dataset of 1000 Hollywood movie scenes and their short descriptions. The scripts are scraped from IMSDb[4]. The scenes are annotated with the four major components of a screenplay: sluglines, action lines, character names and dialogues, described in details in appendix A.4 We introduce a workbench which we call “Kurosawa”, consisting of datasets and a pair of GPT-3 (Brown et al., 2020) models fine-tuned with the said datasets. One GPT-3 model generates a movie plot given a short description of the storyline (15– 40 words), while the other creates a scene based on a short description of the required scene. Importantly, we have provided the “Kurosawa” platform to one of the biggest media platforms engaged in the business of making movies and TV shows, producing music and soundtrack etc.- to help script and content writers from different film industries create new movie plots. Our contributions in this work are as follows: • To the best of our knowledge, this is the first work on generating movie scenes from a scene description. • We create and publicly release two datasets: (a) a parallel dataset of 1000 movie storylines and their corresponding plots, (b) a parallel dataset of 1000 movie scenes and their corresponding descriptions. In (a), we link available movie storylines from IMDb with available corresponding movie plots from Wikipedia. In (b), we link available movie scenes from IMSDb with corresponding descriptions from IMDb. • We manually annotate movie plots according to a 4-act structure which is an extension of the well-known 3-act structure (Field, 1979). Professional scriptwriters from the media and entertainment industry guided us very closely. • We manually annotate movie scenes with four major components of a scene: sluglines, action lines, character names and dialogues, along with a short description of the scene. • We introduce “Kurosawa”: a workbench that consists of multiple datasets and models which can assist script and scene writers in the film industry. This paper is available on arxiv under CC 4.0 DEED license. [1] https://erosnow.com/ [2] https://www.wikipedia.org/ [3] https://www.imdb.com/ [4] https://www.imsdb.com/ Authors: (1) Prerak Gandhi, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, prerakgandhi@cse.iitb.ac.in, and these authors contributed equally to this work; (2) Vishal Pramanik, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, vishalpramanik,pb@cse.iitb.ac.in, and these authors contributed equally to this work; (3) Pushpak Bhattacharyya, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai. Authors: Authors: (1) Prerak Gandhi, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, prerakgandhi@cse.iitb.ac.in, and these authors contributed equally to this work; (2) Vishal Pramanik, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai, vishalpramanik,pb@cse.iitb.ac.in, and these authors contributed equally to this work; (3) Pushpak Bhattacharyya, Department of Computer Science and Engineering, Indian Institute of Technology Bombay, Mumbai. Table of Links Abstract and Intro Motivation Related Work Dataset Experiments and Evaluation Results and Analysis Conclusion and Future Work Limitations and References A. Appendix Abstract and Intro Abstract and Intro Motivation Motivation Related Work Related Work Dataset Dataset Experiments and Evaluation Experiments and Evaluation Results and Analysis Results and Analysis Conclusion and Future Work Conclusion and Future Work Limitations and References Limitations and References A. Appendix A. Appendix Abstract Storytelling is the lifeline of the entertainment industry- movies, TV shows, and stand-up comedies, all need stories. A good and gripping script is the lifeline of storytelling and demands creativity and resource investment. Good scriptwriters are rare to find and often work under severe time pressure. Consequently, entertainment media are actively looking for automation. In this paper, we present an AIbased script-writing workbench called KUROSAWA which addresses the tasks of plot generation and script generation. Plot generation aims to generate a coherent and creative plot (600–800 words) given a prompt (15–40 words). Script generation, on the other hand, generates a scene (200–500 words) in a screenplay format from a brief description (15–40 words). Kurosawa needs data to train. We use a 4-act structure of storytelling to annotate the plot dataset manually. We create a dataset of 1000 manually annotated plots and their corresponding prompts/storylines and a gold-standard dataset of 1000 scenes with four main elements — scene headings, action lines, dialogues, and character names — tagged individually. We fine-tune GPT-3 with the above datasets to generate plots and scenes. These plots and scenes are first evaluated and then used by the scriptwriters of a large and famous media platform ErosNow[1]. We release the annotated datasets and the models trained on these datasets as a working benchmark for automatic movie plot and script generation. 1. Introduction Movies are one of the most popular sources of entertainment for people worldwide and can be a strong medium for education and social awareness. The impact and influence of film industries can be gauged from the fact that Hollywood movies invest *These authors contributed equally to this work 1 https://erosnow.com/ 100s of millions of dollars and often make boxoffice collections of billions of dollars. The first motion picture The Great Train Robbery, 1903— black & white with no sound— was created at the beginning of the 20th century. Since then, the art has gone through several transformations, and now people can instantly access 4K HD movies of their liking on any smart device. Throughout the history of film, two of the contributors to a film’s blockbuster success have been the quality of its plot and the manner of storytelling. The appeal of the movie decreases drastically if the viewers find the plot drably predictable. Writing a creative and exciting script is, therefore, a critical necessity and is extremely challenging. Add to this the constraints of time and budget, and the need for (at least partial) automation in script writing becomes obvious. AI-based story generation has been used before. Based on the engagement-reflection cognitive explanation of writing, the computer model MEXICA (Pérez and Sharples, 2001) generates frameworks for short tales. BRUTUS (Bringsjord and Ferrucci, 1999) creates short stories with predetermined themes like treachery. With the arrival of pre-trained transformer models, automatic story generation has got a shot in the arm. Transformer models like GPT-2 and GPT-3 are extensively used for text generation. These models have shown the capability of generating creative text, albeit sometimes with hallucinations (Zhao et al., 2020). Text generated by these models also sometimes lacks coherence and cohesiveness. On the other hand, template-based models can generate coherent text but lack creativity in generating new characters and events in the plot (Kale and Rastogi, 2020). The process of creating a movie generally starts with an idea which is then used to create a plot which is used as the base to build the movie script (Figure 1). Novel datasets are an important feature of this paper. We closely studied the plots and prompts of movies from Bollywood and Hollywood. Such plots and prompts were scraped from Wikipedia[2] and IMDb[3], respectively. The plots are then annotated using the 4-act story structure- an extension of the well-known 3-act structure (Field, 1979). The 4-act structure and the annotation methods are explained in detail in appendix A.5 and section 4, respectively. We introduce a dataset of 1000 Hollywood movie scenes and their short descriptions. The scripts are scraped from IMSDb[4]. The scenes are annotated with the four major components of a screenplay: sluglines, action lines, character names and dialogues, described in details in appendix A.4 We introduce a workbench which we call “Kurosawa”, consisting of datasets and a pair of GPT-3 (Brown et al., 2020) models fine-tuned with the said datasets. One GPT-3 model generates a movie plot given a short description of the storyline (15– 40 words), while the other creates a scene based on a short description of the required scene. Importantly, we have provided the “Kurosawa” platform to one of the biggest media platforms engaged in the business of making movies and TV shows, producing music and soundtrack etc.- to help script and content writers from different film industries create new movie plots. Our contributions in this work are as follows: Our contributions in this work are as follows: • To the best of our knowledge, this is the first work on generating movie scenes from a scene description. • We create and publicly release two datasets: (a) a parallel dataset of 1000 movie storylines and their corresponding plots, (b) a parallel dataset of 1000 movie scenes and their corresponding descriptions. In (a), we link available movie storylines from IMDb with available corresponding movie plots from Wikipedia. In (b), we link available movie scenes from IMSDb with corresponding descriptions from IMDb. • We manually annotate movie plots according to a 4-act structure which is an extension of the well-known 3-act structure (Field, 1979). Professional scriptwriters from the media and entertainment industry guided us very closely. • We manually annotate movie scenes with four major components of a scene: sluglines, action lines, character names and dialogues, along with a short description of the scene. • We introduce “Kurosawa”: a workbench that consists of multiple datasets and models which can assist script and scene writers in the film industry. This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv [1] https://erosnow.com/ [2] https://www.wikipedia.org/ [3] https://www.imdb.com/ [4] https://www.imsdb.com/

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Using Large Language Models for Zero-Shot Video Generation: A VideoPoet Case Study

"Kurosawa": A Script Writer's Assistant: Conclusion and Future Work

Teleplay.Tech

Too Long; Didn't Read

Boost your HackerNoon story @ $159.99! 🚀

"Kurosawa", A Script Writer's Assistant: Abstract and Intro

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

Advancing Coherent Story Generation: Future Directions for AI Writing Tools

"Kurosawa": A Script Writer's Assistant: Motivation

"Kurosawa": A Script Writer's Assistant: Experiments and Evaluation

"Kurosawa": A Script Writer's Assistant: Limitations and References

"Kurosawa": A Script Writer's Assistant: Dataset

Advancing Coherent Story Generation: Future Directions for AI Writing Tools

"Kurosawa": A Script Writer's Assistant: Motivation

"Kurosawa": A Script Writer's Assistant: Experiments and Evaluation

"Kurosawa": A Script Writer's Assistant: Limitations and References

"Kurosawa": A Script Writer's Assistant: Dataset

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps