I’m afraid it can do that. An AI creates Horror movies

A while back I used a Neural Network to create some anime, then I did the same exercise with Lifetime movies, both had some interesting results. This time, given the season, I decided to go for horror movies. Will we get horrors beyond human comprehension? Are we ready too look into the abyss of an artificial mind? Or would we just get wacky shenanigans? Will these give Blumhouse some ideas? Is this how Blumhouse get their ideas already? Well, let’s find out

How do we go about this?

As in the previous time I did this, I collected data from The Movie Database, who have user curated data, has tags for every genre known to man, good and consise descriptions and, the most important thing for this exercise, a really great and easy to use API.

Thanks again for your help in this incredibly dumb exercise, TMBD!

I queried the TMDB to find all horror movies, I had to limit it to those in English not due to anglocentrism but because I was having some encoding issues and rather than trying to figure that out I just took the cowards easy way out. Script is fairly simple

import json
import requests
import time
#{'genres': [{'id': 28, 'name': 'Action'}, {'id': 12, 'name': 'Adventure'}, {'id': 16, 'name': 'Animation'}, {'id': 35, 'name': 'Comedy'}, {'id': 80, 'name': 'Crime'}, {'id': 99, 'name': 'Documentary'}, {'id': 18, 'name': 'Drama'}, {'id': 10751, 'name': 'Family'}, {'id': 14, 'name': 'Fantasy'}, {'id': 36, 'name': 'History'}, {'id': 27, 'name': 'Horror'}, {'id': 10402, 'name': 'Music'}, {'id': 9648, 'name': 'Mystery'}, {'id': 10749, 'name': 'Romance'}, {'id': 878, 'name': 'Science Fiction'}, {'id': 10770, 'name': 'TV Movie'}, {'id': 53, 'name': 'Thriller'}, {'id': 10752, 'name': 'War'}, {'id': 37, 'name': 'Western'}]}


with open("horror-movies.txt", 'w+', encoding="utf-8") as horror_movies: 
    for page in range(1, 500):
        response = requests.get('https://api.themoviedb.org/3/discover/movie?api_key={api_key}&with_genres=27&with_original_language=en&page={page}'.format(api_key=api_key, page=page))

        data_json = json.loads(response.content)

        for result in data_json['results']:
            horror_movies.write(result['title']+'\n')
            horror_movies.write('\n')
            print(result['title']+'\n')
            horror_movies.write(result['overview']+'\n')
            print(result['overview']+'\n')
            horror_movies.write('\n')
        time.sleep(1)

With this I got a ~3MB file with every horror movie title and overview, which we then feed to the same GPT-2 model that we used previously.

We have to upload the file created (“horror_movies.txt”), download a gpt-2 model (this time I used the 355M one) and then train and generate text. This took about 3-4 hours as I ran the generation on a loop to dump to different files

Results

Well, you would think that the data collection, coding and model training would be the hard part, but to be honest, the wading into all the generated ones and trying to find the gems it’s what takes the most effort. I also need to double check for them that they don’t exist already

Without further ado, here are some of my favorites:

The Exorcistic Lecter

The exorcist Paul Smerconish is sent to perform exorcisms of the wealthy, rich couple in their home, only to learn that the mansion has a secret agenda.

It’s funny because last time we got another “Exorcistic”, not sure what’s the deal with this model and that word. Anyway, this is here mostly because I love the “Smerconish” name and I’m intrigued about a mansion with its own agenda

The Crowbar

A serial killer terrorizes a small town.

I’m astonished this isn’t a real movie from the 80’s slasher craze where you could have a new franchise by just having killer + interesting weapon + gore. I love the straight to basics of this one

The Curse of the Flying Serpent

The story of a flying serpent that is taking over the world.

So “The Flying Serpent” is an existing movie, about Quetzalcoatl killing people, this could be the sequel. Let’s get to it, Guillermo del Toro

March of the Pigs

A demonic sexy clown that comes out of the zoo where he was once confined to terrorize swarms of escapees.

This looks like something Rob Zombie could do, I would also be scared of what sexual awakenings a movie like this could bring

Genus: Hippopotamus.

A plague has decimated the planet’s population. A handful of survivors, with the help of a highly-skilled military team, struggle to survive and to find a way to survive this “End of the World.”

Have y’all read about hippopotamus? They’re strong as fuck and really dangerous. Hell, they’re already invading Colombia. Wake up, sheeple!

The Godfather of Zombies

Professor Tod Brown raises the dead to teach a class at the funeral home.

I’m a sucker for horror comedy, this could really work. It could also work as a commentary on the current state of things, with the worker shortage and having to resort to the undead for it. Someone get Edgar Wright on the phone for the former and Boots Riley for the latter.

The Breakfast Club

A group of college students are invited to a haunted castle to help rescue an unwed teenage girl from a mental institution. Soon, the guests begin to vomit up bloodstained corpses in a bid to get inside the castle.

Ok, hear me out, we reboot The Breakfast Club but as a horror franchise! What do you mean almost every horror movie already has the cliques from the breakfast club?

Dinocroc vs. Superman

Dinocroc vs. Superman is a dark, alternate-reality tale that follows a group of people who have attended a dinosaur zoo. Dinocroc vs. Superman is a true-to-life story about a super-sized crocodile, who is being hunted by a man with a jet pack, and his huge nemesis, who is in the process of partaking in the same action. Dinocroc is determined to prove the existence of the dinosaurs, and to prove that he is the superior species.

Dinocroc vs Supergator is a real movie, this one is just the knockoff… using Superman, who is now a man in a jet pack. Par for the course for Asylum

Tammy and the Tiger Shark

A deadly symbiotic relationship between a beautiful woman and her tiger shark captor unleashes a deadly deadly combination of terror and a savage tiger shark.

I can’t believe Tammy broke up with the T-Rex, what the hell Tammy?

S.N.U.T.I.

A group of young people are stranded in an abandoned town and must survive the night in order to make it out the next day.

The plot is fairly simple, but what does SNUTI stand for? Synthetic Neural Unit To Indoctrinate? Super Negative Ultra Texan Instrument?

The New Mutants

The Rocketeer is a mutant who lives in a lake. His father died when he was six, and now he is trying to find his place in the world.

The New Mutants would have been 1000x times better if it had The Rocketeer in it. Fox really dropped the ball

The Scarecrow Man

A local newspaperman, who has a collection of scarecrows, is looking for a way to get rid of them. When he’s offered one, he decides to use them to kill people in the form of a scarecrow.

I’m not sure this gimmick would work, scarecrows are really flammable. Also, WB might come after us for copyright infringement.

Chowder

A young man is given a mysterious prescription from a mysterious white powder. His only problem is that it might not be what it seems.

So it wasn’t cocaine? Bummer

A Haunting at Love Farm

“A Haunting at Love Farm” is a 1980’s horror film with a heavy Victorian touch, featuring Kirsty and her sister, Dawn, as they go to the country where Kirsty grew up. The film is notable for featuring an appearance from the late, great Woody Harrelson.

RIP Woody Harrelson

Hannibal Lecter: The Vampire

Hannibal Lecter: The Vampire is the first novel in the “Hannibal” series, and is a prequel to the TV series “Hannibal”. It is set in 1930, when Lecter is hired to do research for the upcoming “Hannibal” film, “Exposé”. At a secret location he is set up to do experiments to deliver a cure for an experimental epidemic. The first victim is a young girl, but she starts turning up all over the place.

Is this not a thing because it feels like it could be a thing or at least a fanfic.

Eat the Rain

An American-Japanese chef (Donald Sutherland) and his wife (Adrienne Barbeau) must survive the night in a Tokyo restaurant in order to save the life of the only survivor, a woman who has been suffering from the same strange disease she’s been trying to cure her friend.

The title sounds pretty artsy, like an A24 movie. I can see this working, not sure about Donald Sutherland

By Dr. Mercola

By now you know that sugar is bad for you. But what if you didn’t?
So says the Food and Drug Administration when it approves a new sugar-sweetened medication, a sugar substitute called ‘Stress Superload.’
It’s called Stress Superload because it’s created to help people deal with growing up without a mother.
The FDA has approved an experimental sugar substitute that will make you feel better by making you think you’re growing up. But it’s also going to make you stronger and faster.
It’s like having a baby.
You may have noticed that the holidays are somehow related to the death of Santa Claus.
This is a crossover between “The Crow” and “Vincent.”

There’s so much going on here that I don’t know where to start. Dr Mercola is a real life doctor, he’s a quack that has recently gained notoriety for peddling fake covid remedies. The FDA bit reminds me of The Stuff, although not sure how a sugar substitute can help you deal with “growing without a mother” and also Sant died.

Predator: The Open House

A group of college students visit a haunted house in the woods to interview a local psychic.

After the latest Predator movie, why the hell not?

Frozen Demon

Can a group of young people survive without a drink?

This year? No, I sure can’t

Final notes

This time I got a lot of movies to feed the model, but most of the times I basically just got real movies with different plots. It was hard to get some gems from it. If y’all want to dig into them, here’s a link with the resulting text files. I used a temperature factor of 0.7 which is supposed to scramble the text enough for it to make new things and not that much so it only gets nonsense, but maybe I should have gone lower. Who knows. We got some good ones anyway.

As usual, have fun and post your favorites in the comments!