Incomprehensible (dark fusion jazz music video)

[roxton]

Hi everyone!
Here’s my new music video - a kind of post-Soviet surrealism.

I would describe the music as dark fusion jazz.
Music made with: Arch Linux + Reaper + bass & electric guitars + Drumlabooh + U-he synths + MDA Piano + Vital
Video: Flux-generated with my promts, edited at Kdelinive

[wagtunes]

Hey Peter, great stuff. Sub titles would be nice (I don't speak Russian) but that aside, the music is really cool. I'm curious as to how extensive those prompts for the video were because that is one intricate video. Mind sharing?

[Hipster Bales]

Hi everyone!
Here’s my new music video - a kind of post-Soviet surrealism.

I would describe the music as dark fusion jazz.
Music made with: Arch Linux + Reaper + bass & electric guitars + Drumlabooh + U-he synths + MDA Piano + Vital
Video: Flux-generated with my promts, edited at Kdelinive

Да!, это очень хорошо, чувак! Микс отполирован, а ударные звучат хрустко. (ENG: Yes! It is very good, dude!. The mix is polished and the drums are crispy)

[roxton]

Hey Peter, great stuff. Sub titles would be nice (I don't speak Russian) but that aside, the music is really cool. I'm curious as to how extensive those prompts for the video were because that is one intricate video. Mind sharing?

This music video is edited together from roughly 200 scenes, each lasting between 6 and 30 seconds.
Every scene is generated by a neural network (Flux) based on my script (prompt) - first as a static image, or rather, as many variants of it. I pick the variants I like and then start developing them into video. At this stage I check how much the characters or the environment mutate during the "animation" process. Any variants with too many mutations that can’t be fixed - I discard.
And only once I’ve finally found a sufficiently stable static base for the scene do I develop it step by-step using follow-up prompts.
And so it goes with every single scene. At the same time, a huge number of scenes start drifting away from the original reference image, so I have to endlessly refine and correct them - and every single refinement makes the image quality noticeably worse.
After that, I take all the scenes and edit them together in Kdenlive.
Here’s an example of the script for the intro:

01. Camera is static. A guy and a girl are looking at each other in silence.
02. Camera is static. The guy, in a natural tone of voice, asks: "What should we do? Want me to play the guitar for you?"
03. Camera is static. The girl answers: "Not right now."
04. Camera is static. The guy exclaims: "Oh wait, I’ve got Semiletov’s new music video!"
05.Camera is static. The girl perks up and asks: "Really? What’s it called?"
The guy replies: "Incomprehensible!"
The girl says: "Cool, I want to hear it!"
06. The guy says: "Let’s turn it on right now!" - quickly gets up and rushes to the only videocassette recorder (VCR) in the room next to the TV. He inserts the videocassette into the VCR and presses play.

[roxton]

Yes! It is very good, dude!. The mix is polished and the drums are crispy

Thank you! By the way, the drums were made using Drumlabooh (https://psemiletov.github.io/drumlabooh) - a drum machine I'm currently developing myself.

[wagtunes]

Hey Peter, great stuff. Sub titles would be nice (I don't speak Russian) but that aside, the music is really cool. I'm curious as to how extensive those prompts for the video were because that is one intricate video. Mind sharing?

This music video is edited together from roughly 200 scenes, each lasting between 6 and 30 seconds.
Every scene is generated by a neural network (Flux) based on my script (prompt) - first as a static image, or rather, as many variants of it. I pick the variants I like and then start developing them into video. At this stage I check how much the characters or the environment mutate during the "animation" process. Any variants with too many mutations that can’t be fixed - I discard.
And only once I’ve finally found a sufficiently stable static base for the scene do I develop it step by-step using follow-up prompts.
And so it goes with every single scene. At the same time, a huge number of scenes start drifting away from the original reference image, so I have to endlessly refine and correct them - and every single refinement makes the image quality noticeably worse.
After that, I take all the scenes and edit them together in Kdenlive.
Here’s an example of the script for the intro:

01. Camera is static. A guy and a girl are looking at each other in silence.
02. Camera is static. The guy, in a natural tone of voice, asks: "What should we do? Want me to play the guitar for you?"
03. Camera is static. The girl answers: "Not right now."
04. Camera is static. The guy exclaims: "Oh wait, I’ve got Semiletov’s new music video!"
05.Camera is static. The girl perks up and asks: "Really? What’s it called?"
The guy replies: "Incomprehensible!"
The girl says: "Cool, I want to hear it!"
06. The guy says: "Let’s turn it on right now!" - quickly gets up and rushes to the only videocassette recorder (VCR) in the room next to the TV. He inserts the videocassette into the VCR and presses play.

Wow, I am blown away by the patience you have to do all that.