Projects 16 min read

THE INTRUDER: Plan 9 stalks the AWS console in D minor at 150 BPM

A 3:40 pop-EDM battle-rap. Plan 9 the Bell Labs bunny puts on his headphones at 03:14 UTC, slips into an AWS Availability Zone, and raps every service he passes (Lambda, S3, KMS, kubectl, EBS, Glacier) over a deadmau5-flavoured rework of Bach's Toccata in D minor. Three voice personas (lead Plan 9, autotuned female ad-libs locked to A4, deadpan whisper) plus a Transformer-droid layer in the gaps that delivers the Werner Vogels easter egg, "everything fails, all the time," and the universal devops punchline, "it's always DNS." Five passes from baseline to ship.

THE INTRUDER: Plan 9 stalks the AWS console in D minor at 150 BPM

THE INTRUDER: Plan 9 stalks the AWS console in D minor at 150 BPM

A stick-figure bunny puts on his headphones, slips into an AWS Availability Zone at 03:14 UTC, and battle-raps every cloud service he passes. From cold-start Lambda to Glacier deep archive, the bunny narrates his way through the network while a deadmau5-flavoured pop-EDM rework of Bach's Toccata in D minor pumps in his ears. The hook is Plan Nine in the A-Z, us-east-one. The breakdown is they don't see the bunny, ghost in the etcd. The outro is the most-quoted line in distributed systems and a wink to the universal devops punchline.

Watch THE INTRUDER on YouTube. The engine is open source (GPL-3.0-or-later). The film is Creative Commons (CC BY 4.0).


The brief was simple: take a pop-EDM track in the deadmau5 / Daft Punk language, hand the lyrics over to Plan 9, and let him rap an entire AWS Availability Zone (every service, every CLI, every meme) without naming anything specific enough to be a real threat. 3:40 of music. Half-time pocket on the verses, four-on-the-floor on the drops. Real AWS terminology bent into hard-consonant battle-rap. A hook strong enough to repeat. And bookends with a Stranger Things alien-cricket intro and a Werner Vogels easter-egg outro.

Five passes from v1 to v5. The audio engineering took most of the iterations. The visuals were the easier half; by the throne_protocol pass yesterday the multi-shot cinematography pipeline was already mature.

1. The song, in D minor with a Bach motif and deadmau5 sidechain

The score is a ChipForge composition with nine sections of mostly 16 bars each: intro, verse 1, rise, drop 1, verse 2, breakdown, drop 2, drop 2 reprise, outro. D minor throughout the verses, lifting to D major at every drop, the same parallel-key resolution as PAPER GUNS' Beethoven 5th treatment but used here as the intruder ascending into mastery instead of fate becoming paper.

The melodic foundation is the opening of Bach's Toccata and Fugue in D minor BWV 565: D, C-sharp, D, A. That four-note motif lives in the brass and gets called every four bars. Inverted in verse 2 (call-and-answer with itself). Octave-up in the reprise. Same shape lifted to D major at every hook so the resolution is felt before it's heard.

Production language is deadmau5 / Daft Punk:

  • kick_deep four-on-the-floor on every quarter
  • bass_growl 16th-note pumping, sidechained 4:1 with 110 ms release at 150 BPM (dead-on for the kick recovery)
  • sub_boom long-held foundation under the drops
  • brass_stab carrying the Bach motif with FilterEnvelope-driven attack (base 850 Hz, peak 2900 Hz, sustain 1740 Hz): the "blat" of the brass-section attack
  • pulse_warm, lead_vibrato, arp_shimmer, pad_evolving, saw_dark rounding out the 10-channel arrangement

The master chain is the standard ChipForge stack: parametric EDM EQ → light digital edge → sidechain pump → bus compression → stereo widener at 1.32x → master bus with presence shelf, air shelf, soft limiter at -0.3 dB. The Bach motif and the brass stabs both get ping-pong delay via channel effects, so the four-note motif bounces stereo between the speakers.

204.8s of music. Plus 5 seconds of intro bookend (silence + crickets) and 10 seconds of outro bookend (Werner Vogels card, ASCII bunny, fade). 219.83s total. 2,638 frames at 12 fps.

2. The rap: eight verses, two hooks, an exfil reprise

Verse 1 is the entry: 8 lines × 2 bars at 150 BPM (each bar is 1.6s, 19.2 frames). Half-time pocket. Walking bass under the brass motif. Plan 9 names what he's seeing as he walks the AZ:

Cold-start Lambda, light the fuse.
Slidin' through subnets like a midnight ruse.
V-P-C peering, every door I choose.
S-Three bucket glowin' got me singin' the blues.

The pre-chorus rise builds 8 bars over a saw-dark riser, console warnings ramping (CONSOLE SESSION ANOMALY, READ-ONLY ROLE REBOUND, SIGNATURE MISMATCH, ACCESS-DENIED → ACCESS-ALLOWED), then the drop:

Plan Nine in the A-Z, us-east-one!
Pop that DynamoDB, partition's done!
Route fifty-three routin' me, alias to ghost!
Bezos couldn't catch me with the most!

Each hook line gets a backup-vocalist call-and-response from a female voice locked hard to A4. us-east-one! answers from the right channel. partition's done! answers from the right channel. The lock to A4 (the dominant fifth of D major) is the Cher-style hard-tune signature.

Verse 2 takes him deeper: KMS keys jingling, EBS snapshot in the vest, RDS read-replica primary going dark, Kinesis shard splitting into a stream of consciousness, EKS cluster cracking with kubectl freezing every pod. The breakdown is the whisper section:

they don't see the bunny...
ghost in the etcd...
phantom in the queue...
I am the S-Q-S message that no one ever read.

Drop 2 is the exfil: Glacier deep archive thawed, ElastiCache cycles grooving, Terraform plan applies but the bunny is already gone, stack drift detected but it's just his song. The reprise lifts an octave up: VPC flow logs flippin' to a 404 / GuardDuty barkin' but I closed the door / Plan Nine the bunny they been lookin' for / hop the firewall, baby, see ya in the lore. The outro thins out: sayonara A-Z, leave the door propped, fade, and one final whispered word: out.

Every line is real AWS terminology bent into hard-consonant battle-rap. Cheesy on purpose. Universal/inner-life themes only. No real persons named. No real systems intruded.

3. The voices, three personas plus a Transformer droid

ElevenLabs handles the lead and the call-and-response:

  • og_glenda_bunny: the Plan 9 voice. Carries the eight verses and four hook variants. Same persona that anchored PLAN 9 RAP BATTLE, PLAN 9: EMERGE, THROUGH ME, and THRONE PROTOCOL. Excited intensity 0.9 on the hooks; confident on the verses; menacing whisper on the intro establishment lines.
  • backup_vocal_female (Bella): call-and-response ad-libs in the hooks. Autotuned hard to A4 via rubberband, spit 0.85, blend 0.80, max-shift 12. Locked to the dominant fifth of D major for that brittle Cher-style signature. All 44 cached female clips backed up to *.original.mp3 then auto-tuned in place.
  • the_creator: deadpan whisper on the breakdown bridge, the outro tag, and the Werner Vogels easter egg. Same Joshua-clone persona that closes the other Plan 9 films.

The droid voice is new this pass. The brief was transformer x Skrillex break-beat melodic stepping: a low electronic robotic presence that fills the gaps and chants AWS terminology in tune.

After two passes of overcooked processing (heavy bandpass + 6-bit crush + apulsator wobble + tremolo) the droid sounded like noise. The v5 chain stripped it back: pitch down 3 semitones via asetrate, light chorus (depth 0.55, decay 0.30), light phaser, very light 10-bit acrusher, EQ with body at 800 Hz and presence at 2.8 / 4.5 kHz so consonants are clear, mechanical room reverb, +1.5 dB makeup. That keeps the Transformer flavour and lets the words land.

The droid only speaks in clear empty zones. Four lines total:

  • Region us-east-one. Status nominal.: intro bookend, before the music starts
  • Eventually consistent.: late breakdown, between the last whisper and the drop-2 entry
  • Everything fails. All the time.: outro bookend, after the music fade. The most-quoted line in distributed systems.
  • It's always D-N-S.: final outro line. The universal devops punchline.

The Werner Vogels quote and the DNS line are the AWS insider easter eggs, the "wink to the audience" moments anyone in the cloud-engineering crowd recognizes immediately.

4. The visuals: multi-shot, lip sync, eye contact

Plan 9 Glenda bunny in stick-figure form across nine sections. The cinematography palette swaps each section, mapped to a director:

Section Palette Visual mode
intro bookend kubrick navy empty AZ warehouse, status panel pulsing NOMINAL
intro kubrick navy smoke + neon bars + topology nodes pulsing
verse 1 scott steel server racks scrolling parallax, amber rack glow
rise boyle neon floor grid intensifying, terminal warnings cascading
drop 1 tarantino gold floor grid + neon bars + AWS service badges orbiting
verse 2 fincher green dimmer server racks + floor grid + scrolling kubectl
breakdown burton near-black letterbox 2.4:1 + smoke + dimmed racks
drop 2 boyle neon floor grid + falling 1s and 0s data stream + EBS/KMS/Glacier badges
reprise tarantino VPC/GuardDuty/Firewall badges + bunny ascending
outro kubrick brick firewall (with flames) + bunny exiting

Drop sections hard cut every two bars between eight pre-defined shots: medium body / ECU eyes (4× zoom) / wide stage / head close-up / dutch tilt / repeat. Verses sweep with sinusoidal scale and dutch oscillation plus occasional cuts to head close-ups. The breakdown does an ECU push-in then pulls back into drop 2.

Eye contact with the camera is the default. The bunny is rapping at the viewer. Brief side-glance flicks fire only on the fourth bar of each phrase, and beat-driven punch flicks during drops snap back to center within a quarter beat. Subtle "alive" sinusoidal drift (±0.06) keeps the eyes from looking dead even at full center. The whisper section is the exception: half-closed feel, slow drift, more frequent blinks.

Lip sync is a blend of phoneme-driven and amplitude-driven mouth shapes. The amplitude side comes from lip_sync.build_mouth_timeline analyzing the actual voice mix WAV at 12 fps. The phoneme side comes from phoneme_timeline reading the SCENE_BEATS text to assign visemes per character. The two are blended frame by frame: when amplitude says "speaking", the phoneme shape wins (lip-readable); when amplitude is silent, mouth stays closed; when there's audio without text, amplitude shape wins. The result is more readable than v1's pure-amplitude treatment.

Lighting is built per section: a light_pool orbits around the bunny in drops on a 4-bar cycle; the reprise has two overlapping spots at opposite phases; the rise tightens a single spot from radius 180 down to 100 over its 8 bars. A kick-pulse color wash (additive overlay scaled by exp(-beat_phase * 5.0)) makes the screen "breathe" with every kick at -18 dB strength in drop 1, -22 dB in drop 2. Strobe flashes fire 3-4 frames at the entry of each drop synced to the sub-kick boom in the SFX layer.

5. Blade Runner captions, letter by letter

The captions in v1 were standard lower-third lyric overlays. v5 went kinetic with full Blade Runner / Akira styling:

  • Letter-by-letter type-on reveal synced to voice timing. The cursor block blinks on the trailing edge during the type-on, then disappears once the line is fully revealed.
  • Per-letter position and rotation jitter (±2 px, ±3.5°) seeded by (frame // 6, char_index) so it's deterministic but constantly shifting, handwritten-organic feel without per-frame chaos.
  • 5-pass neon glow halo with 8-direction multi-radius spread, modulated by a per-frame flicker (sin(f * 0.7) * sin(f * 0.21)).
  • Chromatic aberration RGB ghost: red letter offset 2 px left, cyan letter offset 2 px right, behind the main letter. The futurism cue.
  • Horizontal scanline strip behind the caption at 18% alpha: subtle CRT texture.

Color-coded by persona: gold (255, 200, 60) for the lead rapper, bone (232, 224, 212) for the_creator's whisper, hot pink (255, 105, 180) for the female ad-libs (rare; most ad-libs are skipped in the caption layer so the lead's punch word holds on screen).

6. The bookends: Stranger Things in, Werner Vogels out

Front bookend (5 seconds before the music starts):

  • The visual is an empty AWS data centre. Six dim server-rack columns with blinking green and red LEDs (each rack on its own pseudo-random period). A status panel top-right reads AWS · us-east-1 / STATUS: NOMINAL / 03:14:01 UTC / AZ-1A · AZ-1B · AZ-1C with a pulsing green dot. The 03:14:01 timestamp is pi minute, the easter egg buried in plain sight.
  • The audio is Stranger Things alien crickets, no frogs. The earlier passes used synth_pond_ambient.py's default cricket-and-frog mix; the bass-frog ribbits felt wrong in a data center. v5 switched to --crickets-only then layered a metallic shimmer drone (1857 Hz sine through heavy chorus + phaser + echo, the synthwave/Upside-Down shimmer) and a sub-menace drone at 41 Hz underneath. The crickets themselves get chorus + phaser + multi-tap echo for that morphing alien-chirp feel.
  • Title fade: A NAPKIN FILMS PRODUCTION → THE INTRUDER → [us-east-1 · 03:14:01].
  • The droid's intro line (Region us-east-one. Status nominal.) lands at 2.5s into the bookend, in the silence.

Back bookend (10 seconds after the music ends):

  • $ traceroute plan9: six lines of console output, the last one: Connection timed out.
  • $ aws ce get-cost-and-usage: Cost Explorer CLI returning BlendedCost: 0.00 USD. The bunny exfiltrated the kingdom for free.
  • The Werner Vogels quote in a faint outline box: "Everything fails. All the time." (Werner Vogels, CTO, AWS). With (it's always DNS) below it. The most beloved AWS insider one-two punch.
  • PLAN 9 / see you in the lore on a tiny ASCII bunny in monospace.
  • Underneath all of it: the droid speaks the Werner quote at 211s and the DNS punchline at 219.7s, pitched-down, light-processed, clearly intelligible.

7. Five passes from baseline to ship

  • v1: clean baseline. Lyrics + music + voice. Captions in the lower third. Eye animation that scanned everywhere instead of looking at the camera. No droid. No SFX layer. Worked but felt like a draft.
  • v2: full cinematography pass. Multi-shot hard cuts on drops, lip sync, lighting (spotlight + kick wash + strobe), Blade Runner captions, eye animation per section. Visually the biggest jump. Render time stretched to 5+ minutes due to per-letter caption canvases.
  • v3: droid voice layer (13 lines), Stranger Things crickets, drop sub-kicks, transition wooshes, intro/outro bookends rewritten with Werner Vogels and ASCII bunny. The full elevation.
  • v4: female ad-libs autotuned to A4, droid level boosted +6 dB, Plan 9 voice doubler chain (pitch-detuned secondary voice + chorus + extrastereo), 4 kHz presence boost, +0.6 dB body boost. Mix audibility improvements.
  • v5: droid cut from 13 lines to 4 (gaps only: intro bookend, late breakdown, two outro bookend lines). Heavy bandpass + bit-crush + apulsator wobble stripped. Light pitch-down and EQ retained. Words actually intelligible. Final continuity-and-coherence pass.

The droid layer was the longest engineering thread. v3 had it constant and unintelligible. v4 had it louder and still unintelligible. v5 cut the count and the processing: fewer words, every word clear. Sometimes the right amount of a thing is much less of it.

8. Final sound spec

Runtime:       3:39.83 (219.83 s)
Frames:        2638 at 12 fps
Resolution:    854 × 480
Audio:         AAC stereo at 44.1 kHz
Music:         204.80 s (chipforge) — pop-EDM, D minor → D major
Voice mix:     222.33 s (60 ElevenLabs clips, 44 autotuned female)
SFX layer:     219.83 s (crickets, droid, sub-kicks, wooshes, drones)
Music chain:   sidechain pump + EQ + edge + bus comp + widener + master
Voice chain:   HP + EQ presence + comp + doubler + extrastereo + reverb
Final mix:     song_forward profile (voice -2 dB, music 0 dB base)
Verify:        65 of 65 checks passing

9. New scripts shipped this pass

  • scripts/synth_droid_voice.py: generate transformer-droid voice clips. ElevenLabs base voice through pitch-down + light chorus + phaser + EQ + reverb. Configurable per-clip pitch, bit depth, stutter prefix.
  • scripts/synth_intruder_sfx.py: build the unified SFX timeline WAV (crickets + shimmer + sub-menace + sub-kicks + wooshes + droid clips at frame positions). Single ffmpeg filter_complex with adelay-padded streams.
  • scripts/autotune_female_adlibs.py: process all cached backup_vocal_female_*.mp3 clips through autotune_voice.py locked to A4. Backs up originals to *.original.mp3 for reversibility.
  • scripts/_retime_the_intruder.py: one-shot retime helper that auto-snaps SCENE_BEATS frames to actual ElevenLabs clip durations + 0.20 s gap. Resolved a 22 s caption-vs-voice slide between the original musical-bar grid and ElevenLabs's natural delivery rate.

All four are in the Napkin Films scripts/ folder under GPL-3.0-or-later.

Related work

THE INTRUDER sits in the Napkin Films Plan 9 line, same character that anchored PLAN 9 RAP BATTLE (his hardcore-rap debut), THROUGH ME (the EDM-rap meditation in G minor), PLAN 9: EMERGE (the cyber-cathedral emergence), AGENT MODE: OG BOBBY JOHNSON (the OG Bobby crossover), and THRONE PROTOCOL (yesterday's coronation, in F minor at 145 BPM). It is closer in temperament to SATURDAY MACHINE and CANTUS RAVE than to the contemplative line: both are also EDM-forward, both are also rap-anchored.

The biggest production lesson from this pass: less droid is more droid. Constant low-electronic mumbling reads as noise. Four well-placed lines in clear empty zones reads as a character. The Werner Vogels quote at 3:34 lands harder because nothing else is competing for that frequency band at that moment. Restraint is the production move.

The autotune pipeline carried over from THRONE PROTOCOL: the rubberband-based autotune_voice.py first matured on PLAN 9 RAP BATTLE and has now hard-tuned 60+ female ad-lib clips across two films without re-tuning.

The cinematography palette system (Kubrick / Scott / Boyle / Tarantino / Fincher / Burton, etc.) is the same one introduced in PLAN 9 RAP BATTLE and refined through every subsequent Plan 9 film. The Bach Toccata D minor foundation is shared with PORCH POLKA STARGAZER (a different treatment: classical → rock → rave) and the Beethoven 5th approach in PAPER GUNS (different motif, same parallel-key drop trick).

License

THE INTRUDER is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Remix it, repost it, drop it into your own thing, credit "Napkin Films / Organic Arts LLC" and link CC BY 4.0. Engine code (Napkin Films, ChipForge) is GPL-3.0-or-later. ElevenLabs voice audio is generated under their commercial license and is not redistributed in source form. AWS service names and the AWS Cost Explorer CLI syntax appear under nominative fair use as cultural reference for comedic battle-rap. No real systems were intruded upon. No real persons were named. Bach's Toccata and Fugue in D minor BWV 565 is in the public domain. The deadmau5 / Daft Punk production language was studied for spectral analysis only; no audio was sampled, no melody was quoted. Contact: j@organicartsllc.com


On the methodology: This film required twelve production passes and tight human direction across every stage of the pipeline. If you want to understand how AI-assisted development actually works in practice, versus the "vibe coding" you hear about, read the essay.

Produced with Napkin Films and ChipForge, tools built by Joshua Ayson and AI agents at Organic Arts LLC.