Umwelt Drift

Umwelt Drift (3:30) maps organisms’ perceptual fields into glitch-metal liquid DnB and visuals. The latter half reframes the 1441 Kakitsu Incident as choreographed, non-gory ritual—neural cascades.

Comment

Inspiration

I attended ELIM's workshop at AI Festival 2025, where submissions were framed not only as competition but as structured feedback loops—pressure as creative fuel. With the Odyssey deadline one week away, I set myself a challenge: could I translate Jakob von Uexküll's idea of Umwelt (an organism’s subjective perceptual world) into a 21st-century audiovisual experience using AI?

Three catalysts converged:

  1. Recent learning (lyric workshops, DJ Sodeyama on sound precision)
  2. A severe insect phobia requiring non-biological representations
  3. A question: how to show a dragonfly’s data-driven vision without human bias?

The deadline became a forcing function—7 days to test a hypothesis.

What it does

Umwelt Drift is a 3:30 glitch-metal liquid drum’n’bass piece exploring non-human perception through three interwoven layers.
The Umwelt concept is one lens I’m using—not the only one—and this work sits within a broader interest in “in-between” sensory spaces where boundaries blur.

Layer 1: Musical Umwelt

  • Liquid DnB engineered for a floating, unstable sensory state
  • Minimal lyrics with onomatopoeia to bypass linguistic framing
  • Intentional one-beat offsets to create perceptual drift

Layer 2: Visual Sensor Fields — rendered in a post-real realism style

  • Organisms appear as metallic/glass entities (phobia workaround + conceptual fit)
  • Sensory inputs visualized as data-like fields:
    • Dragonfly: Chromatic aberration overlays simulating compound vision
    • Reed plants: Oscillating wireframes showing wind mechanoreceptors
    • Duck family: Thermal halo maps visualizing infrared perception
    • Mycelium: Chemical gradient meshes representing underground signaling

Layer 3: History as Environmental Data

  • The 1441 Kakitsu Incident appears not as human drama but as one ripple in ongoing natural flows
  • Choreographed, non-gory combat rendered as ritual rather than spectacle
  • Violence treated as neural cascade rather than narrative climax

Core question: What if the same moment—a banquet, a pond, a political killing—were perceived simultaneously through a dragonfly’s eyes, reed sensors, and mycelium networks? What would that say about the relativity of “reality”?

How I built it

Timeline: 7 days (Nov 10–17, 2025)

Phase 1: Musical Foundation (Day 1–2)

  • Tool: Suno Studio with my custom prompt generator
  • Chose Liquid DnB to avoid bass-perfectionism sinkholes
  • Built structure + onomatopoeia-heavy Japanese lyrics
  • Exported stems → Ableton for rough mastering

Phase 2: Conceptual Wrestling — The Umwelt Problem (Day 2–3)

Uexküll distinguishes:

  • Merkwelt: The sensory world
  • Wirkwelt: The world of possible actions

How do you visualize a being that senses gradients, polarization, pressure—but not “objects”?

Early attempts failed:

  • Midjourney: too human-centered, too aesthetic
  • Issue: AI models inherit anthropocentric photographic bias

Breakthrough:

  • Treat inputs as raw data streams, not “images”
  • Metallic/glass aesthetics became a translation layer
  • Also solved phobia: machines, not bugs

Frequent prompt refinements: dragonfly POV: faceted chromatic aberration –chaos 40 –style raw → too abstract → added “metallic wireframe overlay, false-color UV spectrum” → lighting mismatch → final: “sensor readout HUD, glitch scan-lines, copper iridescence”

KATHMI’s advice led to switching to nano-banana + Veo3.1 for kinetic signal propagation.

Phase 3: Historical Integration (Day 4–5)

  • Found Kakitsu Incident ideal for exploring scale + perception
  • Rendered warriors as metallic to align with the aesthetic
  • Staged combat as ritual, observed from distant organism POVs

If you’re a dragonfly, a political assassination is just another vibration in the pond.

Phase 4: Editing (Day 6–7)

  • Tool: Final Cut Pro
  • Relied on Claude for rapid, targeted help
  • Discovered FCP’s compound clip + LUT workflow → unified metallic grading

Final stats:

  • 100+ Midjourney generations
  • Numerous video iterations (nano-banana/Veo)
  • 89 final cuts
  • Tokens: substantial, but part of R&D

Challenges we ran into

1. Philosophical-to-Visual Translation

Umwelt is phenomenological. A tick doesn’t perceive “darkness”; it perceives butyric acid gradients.

Failures:

  • Pure abstraction (viewers got lost)
  • Literal simulation (scientifically shallow)

Solution: Hybrid language: sci-fi HUDs, thermal palettes, glitch cues as subtitles for alien sensing.

2. AI Anthropocentrism

Models prefer centered composition and familiar focal logic.

Workaround: “no centered composition, no narrative focus, sensor readout, data visualization”

3. Time Constraint vs Perfectionism

Seven days forced decisions: embrace happy accidents, prioritize coherence over accuracy.

4. Phobia as Constraint

Metallic aesthetic enabled intellectual engagement without visceral discomfort.

Accomplishments I'm proud of

  1. Compressing 100+ years of philosophy into 3:30
  2. A genre blend nobody predicted
  3. Turning a limitation into a signature aesthetic
  4. Building an AI pipeline under intense constraints
  5. Depicting historical violence without exploiting it

What I learned

1. Umwelt as Creative Framework

Forces reevaluation of composition, time, and sensory assumptions.
Prompts become sensor configuration specs, not scene descriptions.

2. Token Waste = R&D

Like film stock—expensive but necessary.

3. Just-in-time Learning

Precision beats mastery when time is scarce.

4. Constraints Become Features

Phobia → metallic aesthetic
Onomatopoeia → universal accessibility
Seven days → ruthless prioritization

Closing perspective

Umwelt Drift is one perspective within a larger, ongoing exploration of perception.
I’m interested in the spaces between worlds—the blurred borders, the layered realities, the non-human ways of sensing that AI helps us translate.

If these directions resonate with you, I’d be grateful if you follow the project as it evolves.

What's next

Short-term (3–6 months): Foundation Building

  • Gather feedback from contests + small showcases
  • Tune “sensory translation” language based on responses
  • Improve infrastructure (local GPU, streamlined Suno → Ableton → FCP pipeline)
  • Develop v2 of my prompt generator with organism-PoV presets

Medium-term (6–12 months): Autonomous Practice

  • Multi-channel installation prototypes
  • Educational POV visualization modules
  • Collaboration with creators studying non-human phenomenology

Long-term (1–2 years): Beyond Umwelt

Umwelt is one lens—future works will explore broader in-between spaces where signals overlap, boundaries blur, and perception becomes layered.

Themes under consideration:

  • Temporal phenomenology
  • Scale invariance
  • And others that will emerge organically through experiments and critique

Success metric: When others can use my workflows to explore their own perceptual questions.

Current identity: Content Syncretist
Current focus: Skills + infrastructure before exposure
Current question: How do we make the invisible perceptible?

Built With

  • flux
  • klingai
  • midjourney
  • nano-banana
  • suno
Share this project:

Updates