Hooroo Jackson

A Tale of Two Firsts

Hooroo Jackson Explores the Two Earliest AI Feature Films in History: Window Seat and THTHNG: Desolation Unknown, and the Battle for the True First

As a lifelong cinephile and a student of film history, I have a built-in passion about the early landmarks of machine-generated cinema. My enthusiasm for entering the nascent AI space was from my long held belief that there was no new avenue to discover, everything was claimed, there was nothing left for the new generations. We were locked out by the older generations. Until AI.

Recent misattributions regarding AI cinema history however have shown that markers in the space are both coveted and contested, as opposed to a natural part of the historic record. 

This article examines the two earliest AI feature films in cinema history: Window Seat and THTHNG: Desolation Unknown, analyzing their respective claims as the pioneering films in the medium.

The overview begins with our AI tools and methodologies:

WINDOW SEAT

July 21, 2023: Wide release on YouTube and Gumroad.
Technical specifications:
* 100 % AI-generated video
* 100 % AI-generated machine acting
* Machine augmented writing

THTHNG: DESOLATION UNKNOWN

April 2023: Premiere screening.
May – August 2023: Two theatrical screenings with Q & A.
September 11, 2023: Wide release on YouTube
Technical specifications:
* Stable Diffusion workflow.
* Largely still image with zooms, pans, shakes, estimates that 10-15 % of the movie is animated.
* AI-generated music.
* Traditional sound but some AI.
* Occasional live action components.
* Real voices processed through AI filters
* Human written script

DEFINING FEATURE FILMS

Both films pass the marker for being films; with the farthest on the spectrum of let’s say unconventional quality, being Warhol’s Empire, an 8 hour film showing nothing but the Empire State Building skyline in still shot; and my favorite to talk about, Derek Jarman’s Blue, a film that is entirely one solid color with experimental sound design. If those are films, then both of our films are more than films.

TECHNICAL COMPARISONS

Visuals

First off, the pure core AI video and performances of Window Seat vs animated slideshow style of THTHNG, including its live action bits, diverts the two films into specific AI milestones right off the bat: hybrid/AI-assisted vs completely AI feature film. Today, there are films coming out with one single AI element that are referring to themselves as AI films, but when we are talking about history, the question of category becomes the very subject.

Writing

THTHNG’s human writing vs Window’s machine augmented writing is the next category. Window Seat was workshopped through LLMs on a line by line, section by section, structural basis, with machines being in charge of every line spoken by the pilot and many of the film’s creative insults. I have always stood by the benefits of writing with AI, and cede no ground as to losing authorship of the screenplay or its status as an auteur work, no different than if I were to have a co-writer in the creative pipeline.

Music and sound

Window Seat uses no AI-music and no AI-sound. The music in the movie is supplemental and only begins 34 minutes in the 61 minute running time. It is important to note AI-sound was not available in 2023.

THTHNG uses AI music and non-AI sound, however it uses some AI sound via a clever workaround: taking relevant sound effects from AI generated music.

Voice-to-voice vs machine voices

This distinction is more straightforward: Window Seat used pure machine acting, while THTHNG used vocal mocap, with AI-processed human voices.

RELEASE DATES

This is a gray area making the determination extremely difficult. In film industry terms, there is no clear marker for what constitutes a release date. Early screenings generally fall into distinct categories:

1. A premiere…

2. Limited release…

3. A festival run…

A festival run will often constitute release due to the significant public and critical reaction that follows. A first screening will, many times, mark a release historically. Wide release generally marks official release today.

Let’s cite historical precedent.

1. “Birth of a Nation” (1915) Case:

• Premiered January 1, 1915 in Los Angeles
• Wide release March 3, 1915
• Historically cited by wide release date despite significant premiere/screening gap
• Set early precedent for wide release as official marker

2. “Citizen Kane” (1941):

• Palace Theater premiere May 1, 1941
• Wide release September 5, 1941
• Historical records typically cite September release

3. “The Jazz Singer” (1927):

• Multiple test screenings preceded its October release, but October is historically cited

However there is a precedent in the other direction where the first screening does constitute release on the public record:

1. “Snow White” (1937):

• Preview screenings in December 1937 mark the historic record despite February 1938 wide release.

2. “Gone with the Wind” (1939):

• Atlanta premiere December 15, 1939, wide release January 1940; historic record is listed as 1939.

Window Seat was published widely, documented, and promoted on July 21, 2023 on Gumroad and YouTube.

(Window Seat’s July ’23 ads)

(Window Seat – July ’23 wide release ad data)

(THTHNG’s historic April 23rd premiere)

(Window Seat’s historic July 21 release)

COMPLETION

In THTHNG’s own words their film was finally completed in September ’23, two months after Window Seat had its completion, world release and critical appraisal. This would mean those early screenings of THTHNG constituted as a work-in-progress screening; the question becomes whether their April screenings constitute an official release.

There is historic precedent for work-in-progress screenings not constituting an official release:

1. “Gone with the Wind” (1939):

• September 9th work-in-progress screenings did not constitute release

2. “The Wizard of Oz” (1939):

• Limited screenings in June/July 1939 with alternate scenes did not constitute release.

3. “Jaws” (1975):

• Early preview screenings in March/April 1975 that led to significant edits did not constitute release

4. “Apocalypse Now” (1979):

• Won Cannes Film Festival with work-in-progress in May 1979
• Official release considered August 15, 1979 in completed state

In total, we have three gray areas:

• Does a theatrical screening constitute official release? Sometimes.
• Does a wide release constitute official release? Sometimes.
• Does a work-in-progress screening constitute official release? 

I would bring forth a fourth gray area. Does an earlier toolkit with a later release constitute as first?

The Tale of the Fox (completed in 30) versus The New Gulliver (1935) case provides a clear historical precedent: although Starevich’s film employed more sophisticated stop-motion techniques and was developed earlier, The New Gulliver’s release in 1935–before Tale of the Fox’s 1937 release–establishes it as the first feature-length stop-motion film with sound.

PUBLIC RECEPTION AND DOCUMENTATION

What distinguishes these two films as historical milestones is their clear public record–unlike theoretical AI films that might exist privately on someone’s computer somewhere, or a 60-minute montage of AI stills existing in the void somewhere; historic precedent indicates there must be release, reaction, and appraisal. 

So I would credit both of us for embracing our films in a very public way. This ultimately is what makes all the difference, as it stands, these are the two earliest works in category and it will be extremely unlikely for another retroactively making the claim without the necessary activity surrounding it.

The marketing elements:

THTHNG Blog Q & A. The reddit AMA. And a long string of social media promotion.

Window Seat had more critical engagement on the public record from RottenTomatoes, a written interview with Mike Haberfelner, a television interview with the Brown & Black podcast, and an episode on the popular FilmThreat video blog.

CONCLUSION

Window Seat and THTHNG each mark distinct milestones in the history of AI cinema.

Window Seat holds its position as the first fully AI feature film in multitudes of categories with its earlier completion and worldwide release, core-AI video and performances, and larger record of critical engagement. On the other end, THTHNG falls in line with first screenings albeit in unfinished state, and earlier toolkit. 

The historic record is not a hard science, however, we can make the following definite claims:

Window Seat is, objectively, the first completed AI feature film, the first widely released AI feature film, the first AI feature film with core AI video and performances, the first feature film utilizing machine writing, and the first AI feature film made entirely by one person. But this does not negate THTHNG as a historic milestone in cinema history representing a pioneering use of AI tools, the first AI-assisted feature film, the earliest AI-toolkit in a feature film context, and something I hold deep respect for: the first public exhibition of an AI film.

Historic Parallels

I am almost seeing this as the feature film equivalent of Le Prince’s Roundhay Garden Scene vs. the Lumière Brothers’ Workers Leaving the Factory. However they invented the underlying technology itself. We may have to go into the 1890s, to Alice Guy-Blaché and George Méliès—Window Seat in this context was the Méliès moment, proving that AI filmmaking was not just a technological novelty but capable of the motion picture experience. This places us in the accelerated timeline somewhere between 1895-1899.

My subsequent film, DreadClub: Vampire’s Verdict, then represents an even greater leap akin to Émile Cohl’s Fantasmagorie (1908), or Winsor McCay’s Gertrie the Dinosaur (1914). In practice, it aligns with the late 1920s and early 1930s, when Disney and Fleischer refined animation into a more cinematic form, prefiguring Snow White (1937). In terms of fluidity, expression, and movement, at points it even surpasses low-budget 1960s anime such as Speed Racer. In the accelerated timeline, that places DreadClub around 1905-1925. Perhaps it lands at Lotte Reiniger’s The Adventures of Prince Achmed (1926) the breakthrough first animated feature film.

The scattered parallels reflect AI cinema’s chaotic and exponential growth. Where traditional filmmaking took 42 years to advance from the Lumières (1895) to Snow White (1937), AI filmmaking has compressed a comparable trajectory into just one year (2023-2024)—one year per decade. As written in my essays, the rapid acceleration of AI cinema points to an endpoint that does not aim to match cinema, but to far exceeds cinema as we know it.

So let it be known: If others films emerge claiming to be “the first AI feature” with little to no existence on the public record, history must recognize that documented release and appraisal are not arbitrary markers—they are the very means by which cinema’s evolution is recorded. Further, if we were to redefine history by what comes later, we must erase Le Prince, Edison, and the Lumières from film history, cinema history then must begin with Griffith, or Chaplin, or Welles. As I argued in AI Films are Films, the logic of denying history’s early milestones would create a never ending chain of ‘true’ first AI feature films. 

History has to start somewhere, and how nice it began not with big money, corporations and film studios, but independent artists.