• Intro
• Theory
• Visualizations
• Methods
• Analysis
• Other

Intro

The term tempo is often associated with music to measure the pace of a song.¹ However, within the context of film, tempo is used to describe pacing and how viewers experience narrative speed (for example, did the film feel long or short). All films have varying degrees of temporality, with certain moments feeling slower or faster to stand out from others. For this reason, tempo is one of the key tools filmmakers use to construct narrative. Examples of filmmakers manipulating tempo include quick cutting during battle scenes to make them feel fast and intense, a lingering shot on someone right before they get murdered to build up suspense, or a song playing over a succession of quick shots (a montage) of working out to represent a training montage.

David Bordwell describes varying techniques filmmakers employ to change narrative tempo: shots with a great deal of fast movement captured (actors running, objects moving fast, disco strobe), intense and dynamic camera movement (whiplash pans, jerky reframings, rack focusing), or quick cutting (making the length of each shot short). However, the main way Bordwell discusses tempo in film is through the lens of average shot length (ASL).²

Theory

How to describe temporality in cinema has long been contested. It has often been left out of conversations of film analysis and theory because of how difficult time is to conceptualize.³ As a response, some theorists have focused on the quantitative aspects of film, such as average shot length, editing speeds, and changes in camera movements. Bordwell primarily uses average shot length (ASL) in his film analysis to discuss the correlation between shorter ASL and quicker tempo in Hollywood cinema over the past century.⁴

Matilda Mroz, however, highlights the qualitative aspects of temporality in film such as a viewer’s life experiences or the time and space a viewer is in when watching a film. Mroz argues that these factors influence an audience’s perception of time. To measure these qualitative aspects of tempo, Mroz explores the roles of duration and rhythm, resonance and uncertainty, affect, sense, and texture.⁵

Figure I is made with qualitative data and can be analyzed through the approach Mroz takes. Inspired by Bordwell’s breakdown of quick cutting to measure tempo, I made another visualization (Figure II) looking at shot boundaries and lengths. My goal in making these visualizations is to bridge the technical aspects of filmmaking and film theory to the experience of viewing one of PTA’s films for a holistic approach to understanding temporality in cinema.

The visualizations were also made to encourage viewers to compare their own experiences of watching PTA’s films to the filmmaking tools he employs. Some questions to consider when using these visualizations could be: Which moments feel faster and slower in PTA’s films? Why do they feel different than the flow of the rest of the film? What is the effect of PTA’s films on his audience? And ultimately, do you think PTA is successful with his application of tempo?⁶

Visualizations

Figure I: Paul Thomas Annotated Screenshots

Open in Full Screen

• How to use Figure I: If you hover your mouse over a tick, the thumbnail for the tick’s screenshot appears along with its screenshot number. The x-axis is the timestamp at which the screenshot was taken during the film (all in sequential order), while the y-axis is each of PTA’s films. The bottom chart works as a slider. You can click and slide your mouse to examine specific moments in the films at a closer look. Using the legend, you can also select a specific film you would like to examine by clicking on the film’s name. To return to the original visualization, double-click anywhere on the visualization.

Note: The last point of every film is the title screen. This isn’t a gap in the film.

Figure II: Shot Lengths and ASL of Paul Thomas Anderson’s Films

Open in Full Screen

• How to use Figure II: How to use the visualization: Using the legend, you can examine one film at a time by clicking the film’s name. You can also move around the graph and zoom in or out (by using a scrolling motion for zooming in and out). If you hover your mouse over a point, you can see the point’s metadata. The x-axis shows the shot number in the order of the film (from the first to the last), while the y-axis shows the length of each shot in seconds. To refresh the graph to its original visualization, double click anywhere on the visualization (if you click on a film, you may need to double-click multiple times before getting back to the original visualization). Note: The last point of every film is the end credits.

Analysis

As an example of the potential ways these visualizations can be used for analysis, I will examine Punch-Drunk Love. Looking at the first half of the film, Figure I displays more breaks (pockets of no screenshots) occurring, while Figure II shows shot lengths are consistently longer (which is an indicator of slow tempo). This makes sense since the tempo of Punch-Drunk Love begins on the slower end. The tempo increases throughout the film and is the quickest at the climax. Punch-Drunk Love begins by following the inner workings of Barry Egan’s life before meeting Lena Leonard. As Barry’s relationship with Lena grows, so does the speed of the film.

An example of when PTA manipulates tempo to display shifting dynamics between Barry and Lena is during their date at the 43-minute (2,608 second) mark. Shot 129 is one of the longest shots of their date. Barry tells Lena about his secret pudding loophole, and the longer the shot continues, the more Barry opens up (emotionally) to Lena. Then, just as Barry finishes talking, Lena brings up an embarrassing story Barry’s sister told her about one of Barry’s outbursts. The viewer can immediately feel Barry withdraw. The shot feels quicker as Barry fidgets, and the tempo of the scene shifts and increases before ultimately, reaching a crescendo in shot 132 after Barry abruptly excuses himself and then destroys the restaurant’s bathroom within the span of fourteen seconds (a relatively quick shot compared to the ASL of this film).

The rest of the film feels increasingly fast. After Barry and Lena’s date, there is a rapid succession of quick shots (and lots of screenshots were captured during this time). The viewer can feel Barry get punched by his sudden and rapidly growing love for Lena through this change in tempo.

Something else to note is that at some points Figures I and II oppose each other. The quicker a film’s relative shot length, the faster the tempo should feel. When sequences involve many shot-reverse shots (typically during conversations) and no new information is being provided to the viewer (characters, angles, etc.), fewer screenshots are taken. This makes sense based on the screenshotting guidelines given.

An example of fewer screenshots being taken during a shot-reverse shot is at the timestamp 1:28:49 (5,329 second mark) in Magnolia. During this scene, the reporter Gwenovier confronts Frank Mackey about his family history. In Figure II, around shot #366, there is no indication that the scene is slowing down; rather, the shots following #366 are all much shorter than Magnolia’s ASL. However, the scene itself feels painfully slow as Mackey’s past comes to light. The rhythm of the film stalls as the viewer awaits Mackey’s secret to be revealed, like waiting for a response with bated breath.

This is why, although shot length is a good tool for viewing filmmaking techniques, quantitative data isn’t enough to understand the tempo of a film. Instances such as above-mentioned example are why it’s useful to have access to both visualizations when measuring tempo. Neither quantitative nor qualitative data alone is enough to fully grasp tempo in a nuanced way.

These visualizations can also be used to look at the pulses of PTA’s films. For example, why do There Will Be Blood and The Master almost look rhythmic with the number of long shots and where they appear in the films, versus Licorice Pizza appearing more level. You can also look at how PTA’s films have changed over time: have the ASL of his films gotten longer or shorter? Are PTA’s films all unconventional (in length/ASL/temporality) or does he employ established filmmaking techniques? Looking at Figure I, you might ask: Why does One Battle After Another have over three-hundred more screenshots than Magnolia if Magnolia is PTA’s longest film? Are the pockets of no screenshots being taken a rhythmic lull in each film or are they sporadic? There are many questions you can pose, and hopefully, Figures I and II are helpful for finding answers.

Methods

Figure I: Paul Thomas Annotated Screenshots

Figure I is a visualization of all screenshots taken by the PTAnnotated team to break down the inner workings of all ten of PTA’s films. For reference, each film has somewhere between 359 and 1144 captured screenshots. These screenshots were taken manually over the span of two years (2024-2026) by PT-Annotated team members under the following guidelines:

• Capture a screenshot any time the focus of the frame changes (shows something new).
• Screenshot title cards/title screens at start of film (even when it’s a blank screen).
• Don’t take a screenshot every time the camera switches for shot-reverse-shots during conversations, only when something new occurs within the shot-reverse-shot scene (i.e., change in angle, new character, etc.).
• Too many screenshots is better than not enough. These screenshot guidelines essentially demonstrate what the PTAnnotated team deemed important information for comprehensively analyzing PTA’s films. It’s important to note that not all captured screenshots are uniform, as multiple team members took screenshots over the years. This is to say, the screenshotting process was a subjective one that probably has some slight inconsistencies. Thus, this visualization mirrors the different experiences viewers have while watching PTA’s films through the use of qualitative data.

Gathering the Data for Figure I

I sorted through the PTAnnotated metadata for the process of capturing screenshots and cleaned it up. I only kept the necessary data for my visualization, which included: the screenshot IIIF manifests, the timestamp of screenshot/scene number, and the screenshot name in the PT-Annotated database. I then had to convert all the screenshot timestamps to seconds to make it easier to graph them as data points. Overall, this process was relatively simple.

Figure II: Shot Lengths and ASL of PTA’s Films

Figure II’s visualization includes a scatter plot with shot lengths and a bar chart with the ASL of each film. This visualization serves as a quantitative approach to understanding film tempo.

Gathering the Data for Figure II

The process for gathering data for Figure II was more complicated. I used the application MakeMKV to decrypt all ten of PTA’s films and then produce MKV files.⁷ After procuring the MKV files of each DVD, I then converted the MKV files to MP4 files because MKV files are not compatible with Distant Viewing Explorer’s open-source code created by Taylor Arnold and Lauren Tilton. To convert MKV files to MP4, I used FFmpeg. Once I had the MP4 files, I ran each one (one at a time) through Distant Viewing Explorer’s open-source code in JupyterLab. This process takes a very long time (around forty to seventy minutes per file).

Distant Viewing Explorer detects shot boundaries (when one shot is cut/transitions to another shot) of videos. The open-source code uses dvt (distant viewing toolkit) to open, read, and provide metadata for video files. Once the video has been opened, the shot boundary detection code runs through the video looking for changes in pixels to mark when a shot cuts/transitions. Once the code finishes running, the final step was making a data table which displayed the timestamps and frame numbers of all the shots. For my own data, I had to calculate shot lengths from the given timestamps and then calculate ASL from the calculated shot lengths.

View code and data for this visualization

Other Applications

Beyond examining one of PTA’s films, these visualizations could lend themselves for cross-examining PTA’s work with other famous directors, with films being produced now, or further temporal examination within film. Other research projects that have conducted inquiries into the role of tempo within film include:

Felix Broberg and Andreas Panagiotidis’ research aims to explain the role of tempo in film, and they conduct a study on action films. The conclusion the essay reaches is that when shot lengths are shorter, the viewer will perceive the film as having a higher tempo.⁸

Brett Adams, Chitra Dorai, and Svetha Venkatesh conducted a study to demonstrate how tempo is used by filmmakers. The study tests the ability of machine learning to see if it is able to gain important information about the film. The study was successful, and thus the authors conclude that shot length and motion can be used not only to change a film’s tempo but also to convey meaningful information to viewers.⁹

Bibliography

• Adams, Brett, Chitra Dorai, and Svetha Venkatesh. “Role of shot length in characterizing tempo and dramatic story sections in motion pictures”. Proceedings of the eighth ACM International Conference on Multimedia (2000): 353-355.
• Bordwell, David. The Way Hollywood Tells It: Story and Style In Modern Movies. Berkeley: University of California Press, 2006. https://hdl.handle.net/2027/heb08199.0001.001.
• Broberg, Felix and Panagiotidis, Andreas. “Impact of shot length and motion on cinematic tempo.” KTH Royal Institute of Technology, 2022. https://kth.diva-portal.org/smash/get/diva2:1704346/FULLTEXT01.pdf.
• Costa, D. “tempo.” Encyclopedia Britannica, April 9, 2026. https://www.britannica.com/art/tempo-music.
• Mroz, Matilda. Temporality and Film Analysis. Edinburgh: Edinburgh University Press, 2012. [http://www.jstor.org/stable/10.3366/j.ctt3fgs7c]
  (http://www.jstor.org/stable/10.3366/j.ctt3fgs7c).
• Taylor, Arnold and Lauren Tilton, Distant Viewing Explorer, 2025. https://distantviewing.org/dvexplorer.
• Taylor Arnold, and Lauren Tilton. Distant Viewing Computational Exploration of Digital Images. Cambridge, MA: The MIT Press, 2023. https://doi.org/10.7551/mitpress/14046.001.0001.

Notes