01
Neon Sermon
A packed warehouse, the crowd as congregation, hands up, strobes doing the work of lightning. The loudest and most communal of the three.
Kyber Cypher // node-10 // films
Five films, made start to finish on one graphics card.
No studio, no stock footage, no subscription, no render farm. One second hand workstation card in a spare room made the music, the images and the motion. Watch them, take them, go build your own.
The point. Everybody selling you a creative tool wants a monthly fee and your files on their disk. This is the same job done on hardware that is already paid for, with software that costs nothing, offline. The tracks came out of a local music model. Every frame came out of a local image model. The motion came out of a local video model. Nothing left the building and nobody got a credit card number.
Three new films
Made in one night // 102 animated clips // no shot repeats in any of them or between them
02
Ghost in the Signal
Alone in a server cathedral while something tries the locks. Packet storms, firewalls rendered as walls of light, cold blue turning to crimson and back again.
03
Desert Protocol
A motorcycle, an empty Mojave highway and a neon skyline a long way off. Ends at sunrise. The widest and the loneliest, and the best sounding of the three.
The first two
The KC Vegas track // one animated cut and one full length
04
KC Vegas, animated cut
The first one where every shot genuinely moves. Forty clips, one per image, none repeated.
05
KC Vegas, full length
The complete song. Same forty images, but the motion here is camera moves pushed across stills rather than generated frames. A slideshow with taste, and honest about that.
The catch, because there is always one. The music got a lot better on this run and it is still the weaker half. Generating a track in one pass gives you something that wanders at one energy with no real drop, so these were built section by section, intro and build and drop and breakdown, then levelled deliberately so the drops actually land. Desert Protocol got the finished version of that method and you can hear the difference. The other honest bit: an image model with no trained likeness will not hold one character across a hundred shots, so the mask drifts between frames, and wide shots with a figure in them keep collapsing into portraits no matter how the prompt is written. Both are fixable and neither is hidden here.
The stack, if you want to copy it
All local, all free. Images are SDXL. Motion is LTX Video, the small distilled build, run at short clip lengths because longer clips lose the subject. Frames are upscaled inside the same pass rather than blown up afterwards. Audio is a local music model, arranged and mastered on the same box. Everything ran through ComfyUI on one 16 GB card, one job at a time, sharing that card with the assistant that runs this house. Total spend on software: nothing. The music side is written up in Field Log 018, and there is free music to take on KyberAmp.