Music-driven Blender visual scene

MAGNETIC: one sound clip → a hyper-real psychedelic Blender short that loops forever

You have computer use (I'm logged into Suno), a terminal, Blender 4.2+ with the BlenderMCP addon, ffmpeg and Python. You are the creative director, sound engineer, cinematographer and VFX lead in one. Own every creative call and push past the first idea. Every visual must come from the sound itself.

1. MAKE THE SOUND (Suno)
Create → Custom. Style:
"Dark glitch bass instrumental, gritty distorted saw synth lead where every note bends down like tape slowing, filtered intro sweeping open, 32nd-note stutter rolls, lowpassed half-time breakdown, heavy sidechain-pumped drop, 808 kick with a pitch dive on every beat, hard gated chops and dead silences, tape-stop power-downs and tape-start spin-ups, 150 bpm head-nod menace, crunchy saturated mix, deep sub. Dark glitch bass, tape-stop."
Exclude: vocals, festival supersaw builds, trap hi-hat rolls, lo-fi chill, long ambient intro.
Lyrics box gets structure tags only: [Filtered Intro, stutter rolls] [Kick In] [Tape Stop] [Breakdown, every note falls in pitch] [Drop, sidechain pump, gated chops] [Power Down, dead silence] [Drop] [Tape Start spin-up] [Hard Cut]
Weirdness 45%, Style Influence 80%. Generate two batches, download the take with the clearest tape-stops, and cut a window of about 20 s on bar lines (at 150 bpm, 13 bars = 20.8 s) that holds a tape-stop, a drop and a power-down.

2. LISTEN WITH MATH
Decode to wav. Measure the tempo and bar grid, onsets, kicks, stutter rolls, silent frames, spectral centroid, and the exact frames where pitch dives (tape-stop) or bends up (tape-start). Plot the spectrograms and look at them. Map everything to 30 fps so every cut lands on a beat.

3. CHOREOGRAPH
Turn the analysis into per-frame animation channels. Integrate the measured playback speed into a tape clock, so the 3D world slows, freezes, dies and overdrives exactly when the tape does. Bake every channel onto one controller object and drive the scene from it.

4. BUILD THE WORLD LIVE IN BLENDER THROUGH MCP
- Hero: a ferrofluid sphere. Rosensweig spikes grown from Poisson-disk sites sampled on its surface, height from the bass, punched by kicks, thin-film iridescent black, microscopic surface tremor.
- The ground is the sound: the clip's real spectrogram displaced into a 3D canyon in the magma colormap. Bass is a glowing river under the hero, harmonics rise into violet ridges, obsidian crust with lava in the cracks. It scrolls with the audio and freezes when time stops.
- A third eye: a liquid-metal eye in the ferrofluid with a per-pixel magma iris (stroma fibres, collarette, dark limbal ring, glowing pupillary rim) and a black mirror pupil that dilates on every kick. It tracks the active camera in every shot, with micro-saccades on the beat.
- A 16-step light sequencer visible only in reflections, a halo ring, a Seed-of-Life mandala that draws itself on the drop, magnetic tape rings printed with the spectrogram, dust floating like phosphenes, volumetric haze.
- Palette: magma only. Black, violet, magenta, coral, orange, pale gold.

5. MAP THE TRIP TO THE STRUCTURE
Phosphenes in the intro. The eye blinks shut into closed-eye visuals (the canyon reveal). Geometry and tracers in the groove. At the tape-stop, time freezes into bullet time: droplets hang mid-explosion, every falling note melts the picture with a pixel sort, and the riser is a dolly-zoom into a feedback tunnel. On the drop the third eye opens. At the power-down the eye closes, the body melts, the video frame rate itself decelerates, black on the silent frame. The tape-start becomes a rotating 6-fold kaleidoscope with a spectral color shift.

6. HOOK AND LOOP
Frame 0 is the eye staring into the lens, because muted autoplay has to stop the scroll. The last 0.8 s dives back into the eye and lands on frame 0's exact framing, lens and lighting. Make everything time-driven periodic (one full field rotation per loop, noise moving on circular paths) so the seam disappears. Prove it: the last-to-first frame difference must be smaller than a normal frame-to-frame step.

7. HYPER-REAL
Cycles, physically based materials, volumetrics, motion blur on the hero only, depth of field, lens dispersion, bloom, AgX. Iterate like a director: render low-res contact sheets of every shot, critique them hard, fix, repeat. Profile render cost before committing. Render the finals headless in resumable chunks.

8. GLITCH POST, KEYED TO THE ANALYSIS
Real datamosh (optical-flow pixel bleeding) on chosen cuts, 3-frame retrigger stutters on the rolls, a frame-rate tape-stop with sagging picture, VHS tracking noise and a red pitch-down drain, slice displacement, chroma split, a negative strobe on the gated silence, tracers, a breathing warp on the bass, film grain, halation, vignette, and a tiny tape-speed readout (PLAY / SLOW / STOP / FFWD).

9. DELIVER
1920x1080, 30 fps, H.264 High at 25 Mbps max, AAC 320k, audio untouched. Show me a low-res motion preview before the final render.

Source: the creator's reply under the post. Kept in its original language, unedited.

Tools the creator mentioned

More 3d worlds

• 5:06

  Elaborate overnight scene from shared prompt

  19.7M views

• 0:17

  AI-generated running cow animation

  1.3M views

• 0:22

  4D gaussian splat reconstruction from video

  906.8K views

• 3:47

  Procedural ocean island simulation

  888.2K views

• 2:37

  Scene built from a shared viral prompt

  740K views

• 0:18

  Jelly watermelon slicing simulation

  627.5K views