This story is over 5 years old.


This Programmer Will Show You How To Make Instant Movie Supercuts

Art hacker Sam Lavigne told us about his app Videogrep—a supercut shortcut that make YouTube gold without any hassle.

The supercut has become the Internet’s practice of choice for humiliating lazy, cliche-ridden, or repetitive media. We’ve been in love with the genre ever since someone lampooned the outrageously inaccurate portrayal of imaging tech in cop shows ("Zoom in on the reflection" gets recycled a lot), and they’ve only gotten funnier since. Entire YouTube channels have been founded on such videos. The creators invest hours and hours of painstaking research and editing into each supercut in return for sweet, sweet likes and views—and many have earned millions of them. However, one clever coder is changing the game—as they often do—by algorithmically streamlining the entire supercutting process.


Sam Lavigne is a computer artist and game designer who’s at the cutting edge of creative computing. He’s run a technological experimentation blog called ‘Work In Progress’ since the fall of 2013, created such crazy contraptions as a code to transform a camera into a musical instrument, and even made a fully-automated stop and frisk robot. Now he has written a code called Videogrep, an app offering to the supercut biz. And the results are phenomenal.

Watching a cycle of speech lead-ups that never reach climax in ’Total Silence’ is awkward, frustrating, cringe-inducing, and perfect. We haven’t even been able to watch the whole thing at once—only a computer could craft something like that.

Lavigne spoke to The Creators Project about Videogrep, the art of supercutting, and the limitations—and potential—of procedurally generated art.

The Creators Project: Where did you get the idea to create a super-cutting super-code?

Sam Lavigne: Lately, I've been interested in procedurally generated text—for example, I made a Python script that transforms literary texts into patent applications, and I've also been experimenting with procedurally generated cinema. I started to create a system using surveillance cameras that acts as a director and an editor, automating camera movement and determining when to cut. At some point, I decided I should try to combine those interests, and use some of the techniques from my text experiments to generate edited videos.


How does the program work? 

Videogrep is a python script that searches line by line through the text of subtitle files for a given phrase, finds the time stamp for each line of dialog that matches the search, and then uses the amazing MoviePy library to create clips and stitch a new video together. It can handle multiple video files.

The basic search takes any regular expression, but you can also give it a hypernym (a semantic category, like "liquid" or "tool") or part-of-speech tags (shorthand for different grammatical categories). For example, "NN" is a noun and "JJ" is a verb, so you could search for all lines of dialog that contain a verb followed by an adjective. For real example see my TED supercut.

Using the hypernym search, you might, for example, make a supercut of all references to "body parts" in Cronenberg movies (I haven't actually done this yet but probably should). Both of these features make use of Pattern, which is a great natural language processing library.

What was the most challenging part of creating it? How long does it take?

Creating it wasn't actually that difficult, since Zulko's MoviePy library and the Pattern library do most of the heavy lifting. The harder part is figuring out what to make with it that's actually interesting. That is, what source material and search terms to use to make compelling supercuts. I spent a long time downloading different movies I love or hate, mass downloading videos from the White House Youtube channel and from TED, and then experimenting with different search phrases.


I experiment a lot, so the time varies. The TED talk supercut took a while because I had to write a script to mass download TED talks and their subtitle files, then did some analysis to find the most commonly used grammatical structures, and then just played around until something felt right. The "Total Silence" video on the other hand basically made itself.

Supercutting—a long, and tedious artform—has been a certain YouTube staple now, for years. How do you think your code affects the people who have been doing it 'the hard way' for so long?

I think that well crafted video compositions still need to be put together by hand—Videogrep can't and isn't intended to replace a human editor. For example, you wouldn't be able to make something like this or this with it. The greatest strength and weakness of procedurally generated work is that the computer has no concept of meaning. When the computer speaks, it doesn't know what it's saying; when it edits, it doesn't really know what it's seeing. It's free from language, which means it's unrestrained in its ability to make wild associations. In this sense, I think Videogrep could be a strong tool to explore and experiment with video quickly, to help find surprising and exciting juxtapositions.

Eventually, of course, all art will be generated by computers, with the only uncertainty being if they will prefer to make poppy trash art, or obtuse high art.


Of course! Give me your wildest imaginings of what completely computer generated, autonomous art might be like.

It's obviously impossible to see beyond the event horizon of the autonomous art machine singularity/apocalypse, but I would speculate that the advent of completely automatic art will give rise to automatic art history, wherein intelligent critique machines hold endless academic conferences, subtly perform character assassination on each other, and vie for rapidly vanishing procedurally generated tenure track positions.

And what are you working on now?

Right now I'm taking a pause from generating new videos to fix up some bugs in Videogrep. I'm also starting work on a collaborative novel about data entry jobs, using a methodology I'm calling "Object Oriented Fiction" or "OOF" for short.

For more crazy experiments beyond the normal realm of tech, check out Lavigne’s personal website here, his experimental blog here, and his Youtube and Vimeo channels here and here.

Image via


See A Supercut Of Wes Anderson's Symmetrical Center Shot

Supercut Film School: Q&A With Elusive, Anonymous Video Essayist kogonada

Twitter's "Art Assignment Bot" Is Generating Prompts For Your Next Masterpiece

Beautifully Simple Code Generates Realistic Terrain With Fractals