These were in the training data:

1917
300
Alien
Amelie
Arrival
Baby Driver
Blade Runner 2049 [Orange/Blue]
Drive
Dunkirk
Ford V Ferrari
Harry Potter and the Deathly Hallows
HER
Interstellar - [Cornfield/Miller's Planet]
John Wich Chapter 2 [Blue/Night]
Jojo Rabbit
Joker [Cold/Warm]
Knives Out
La La Land [Day/Dawn]
Mad Max Fury Road
Midsommar
MoonLight
Moonrise Kingdom
Nightcrawler [Day/Night]
Once Upon a Time in Hollywood
Parasite [Basement/Rain]
Pirates of the Caribbean On Stranger Tides
Prisoners
Seven
Skyfall
Star Wars The Rise of Skywalker [Desert/Blue]
Tenet [Yellow/Outdoor]
The Dark Knight [Night/Night Blue]
The Godfather [Dark/Outside]
The Grand Budapest Hotel [Bonus]
The Lord of the Rings - The Return of the King
The Matrix
The Revenant
The Social Network
The Wolf of Wall Street [Indoor/Outdoor]
Whiplash

A Prayer Before Dawn
Bad Boys for Life
Bill and Ted Face the Music
Birds of Prey
Bloodshot
Color Out of Space
Da 5 Bloods
Devs
Extraction
Greyhound
I Know This Much Is True
Mulan
No Time To Die
Project Power
Sputnik
Tenet
The Batman
The Boys
The Gentlemen
The Grudge
The Invisible Man
The Lodge
The Old Guard
The Outsider
The Witcher
Underwater
Wonder Woman 1984

but it should work with more, eg David Lynch cinematic LUT, Wes Anderson cinematic LUT, Quentin Tarantino cinematic LUT all give kinda consistent results even tho they weren't specifically in the dataset

thanks! it should be a similar effect, i used the same training data for both

well if you pay for my time and rent costs and then wait for a few weeks until the training completes (or several trainings if you count the failed attempts), then sure! otherwise i'd say this is good enough for its purposes 🤷‍♂️

@CCPC yeah i know, i used bulk LUTs previewer to generate the training images. you need to have the LUT files for that though. what I was hoping for is that the model understands the concept of a LUT by the vast variation of them, and then you're able to call for anything you can imagine and see how it handles that.
so I don't get your point

not all LUTs are designed to add color pop, they're mostly changing the contrasts, color palettes to a specific vibe. the naming can be a bit tricky sometimes, eg a LUT named Romantic Mood might change the tones to a more orange/pink palette, less contrasty, more dreamy, but sometimes it gives you unexpected results if the creator "misnamed" it, or when used on a different picture/scene.
for this experiment I just threw in 800 LUTs with 6 images (4800 img dataset) with their original names appended to the image captions, so it might add different variations for the Dramatic keyword as there were different LUTs named Dramatic Action Look #X

I might try a more curated set sometime later but it's a pretty time-consuming project :)

but it kind of caught on the meaning of LUTs, and also changes the surroundings a bit to better match the prompted vibe, so I found this version surprisingly useful already

some examples here: https://eldamar-studio.com/product/8000-wonderful-luts-and-lightroom-presets-bundle/

on the race driver example you can see a bit more dramatic contrast, with an action movie-like color scheme, same with the winnie pic, except it chose a more horror-like color-palette