We Will Never Die (Music Video)

We Will Never Die

March 2026

Title shot of the 80s/90s dorm-room set, with a TV displaying lyrics

Dorm-room set detail with era-typical props

The dorm-room desk, with a Macintosh computer showing song lyrics

Final shot of the music video, with props levitating in the dorm room

Role: Concept, 3D & Animation, Editing
Tools: Blender, BAT add-on, After Effects, Premiere Pro
Time Spent: 3 weeks
Team: Solo (B.A. thesis)

"We Will Never Die" is a music video I made for my Bachelor's thesis, built to show off my Blender add-on, Better Audio Tools (BAT). Where my other thesis piece, Resonance, is a general visualiser for any song, this one is fitted to a single track by Päg Hagström. Much of the camera work and prop animation follows its rhythm.

Art Direction

The whole video lives in one 80s/90s dorm room, staged like a physical memory and filled with era-typical props: a Macintosh, a CRT TV, a synthesiser, an old toy robot. A subtle grade with grain and colour fringing makes it read like aged camera footage, warm and worn rather than clean and digital.

Wide shot of the dorm-room set from the We Will Never Die music video — Dorm-room set with the TV showing lyric motion graphics

Close-up of an electric guitar's pickups and strings in warm light — Guitar close-up, with audio-reactive strings

The dorm-room set as grey untextured geometry: desk, chair, CRT TV, synthesiser, plants and shelves — Solid view of the scene

Better Audio Tools

The video runs on Better Audio Tools, an add-on I developed that brings audio analysis into Blender so a song can drive animation and shading directly. I created it largely with Gemini, as an experiment in how far AI can help an artist write their own tools. The add-on runs locally and splits a track into stems (vocals, instruments, bass and drums), each of which becomes an animation source.

Blender's Video Sequence Editor with five audio waveform tracks: the full song and its four stems — Separated audio stems in Blender's Video Sequence Editor

Under the hood, BAT runs the separation with the HT Demucs model and lays the stems out next to the full mix as tracks in Blender's Video Sequence Editor. Because Blender's node editors can't read audio, each stem's analysis is packed into a 32-bit image texture: the colour channels hold spectral data, the alpha channel stacks one-dimensional features such as loudness, and every pixel column is one frame of the song. Long songs continue across UDIM tiles of up to 4,096 pixels, so the texture never exceeds Blender's size limit, and custom decoder node groups read the data back in both Geometry Nodes and the Shader Editor.

Four stacked texture layers in red, green, blue and grey, each holding a different audio analysis — The RGBA layers of a BAT data texture

Diagram: a song waveform above two 4096-pixel spectrogram texture tiles, 1001 and 1002 — UDIM tiling: a 3:15 song split across two 4,096-pixel texture tiles at one pixel per frame

Animation

Almost everything that moves is driven by those stems. The toy robot raises its mic as the vocals come in, the synth keys follow the bassline, the guitar strings ring on their notes, and the camera rides a rhythmic sway on top of the hand-animated moves.

Blender interface: the dorm room seen through the camera, a CRT TV showing lyrics, with outliner and timeline around it — Manual camera placement on the Blender timeline

Blender node editor beside a close-up of the synthesiser keyboard — Node setup for one synth key, which follows a single note

Blender node editor beside the toy robot holding a microphone — Node setup for the robot raising its arm on a vocal trigger