Words. Instruments. Arrangement. Mix then Master.
The songwriting comes first: the words, and the rhythm they inspire. Then piano, voice, sound design, samples, generation, and arrangement come into the session. There is AI all throughout, interwoven, in the open, on purpose. It analyzes and reviews and proposes and automates. We decide what to keep.
First thing is first
Lived experience, thought experiments, emotional catharsis, praise and worship, research findings, and world events land on the page, sussing out rhythm before any word gets sung or beat gets written. Sometimes it starts with a simple rhyme that won't sit still. Sometimes with an idea that needs to be fleshed out. When a piece starts out as an instrumental instead, it starts at the piano. Guitar also finds its way in, followed by a cacophony of instrument timbres.
Anything that can make sound can be an instrument
Ableton Live 12 and Scaler 3 hold the first sketch and help settle the harmony and structure. Piano, guitar, voice, NI Kontakt, Massive X, Serum 2 build many worlds around the words. Suno and other AI systems come in as instruments in the rack: voice modeling, phrasing, cadence, trying different words and vowel sounds, take after take, vocal comping. Some voices are our own, sung loudly into small rooms. Others are carefully modeled and meticulously directed.
Possibility becomes composition
The arrangement moves back and forth between Ableton Live and Suno Studio as the song takes shape. Generative AI can't finish a record. It provides raw material: performance iteration, happy accidents, parts to dissect, examples of what not to do. Stems process through Live, Moises Pro, or more recently, Suno's generative stem separation. Parts get recombined, performed again, pulled to MIDI and rebuilt, replaced, or left exactly as they came. This is composition, the way producers have cut and comped multitrack records for decades.
One session, one set of decisions
The final arrangement lands in the DAW as a curated multitrack session. Neutron 5 on every channel. Serum 2, Massive X, Kontakt, performed parts, generated stems, and a smattering of Splice samples (when needed) sit in one custom mixing template that streamlines the road to the MASTER. Claude, wired into the session through Producer Pal, gives another set of "ears" (analysis) and assists with cleanup, automation, and checking against the bias we build toward our own mixes.
＋ Session notes: mix architecture
Ableton 12, iZotope Neutron 5 on every track, dual Unmask architecture on the music bus to protect vocal and drums from inter-source masking. Sidechain compression tuned in hierarchy: attack first, ratio second, threshold last. DRUM BUS uses the Dynamic Mix Bus Neutron preset to replace EQ + Glue. Gullfoss tested and bypassed on chamber-acoustic material because the Neutron stack already covered the perceptual work. ShaperBox 3 with rhythmic Drive on the hi-hat for the clarity-shine on top.
The final file, delivered
Per-track masters come first: each runs through Ozone 12 alongside a specially selected reference track, tonal balance weighed by ear. Then a cohesion pass, so the separate tracks land as one body of work with synergy. Everything up to the final file is craft; everything after is distribution. All of it honed on a humble Mac Mini in a tiny home north of Austin.
＋ Session notes: mastering targets
Ozone 12 Master Assistant, targets chosen per project (Indie Pop and Orchestral among them). Loudness set track by track, not to a single number: measured masters run roughly -11 to -15 LUFS integrated, with the bassiest, hardest-hitting track sitting hottest by design. True-peak ceiling held at -1.5 dBTP, Maximizer in IRC 5. A mid-master verification audit catches any plugin that silently deactivated.
Alignment, not defensiveness
Red Plastic Records releases include a plain-language account of where generative AI entered the process and what stayed human-authored and directed. Where distributors' metadata supports it, those contributions are submitted for display in platform song credits. ASCAP, BMI, and SOCAN aligned in October 2025 on registering partially-AI compositions. This catalog and label exist exclusively in this lane.
The work is documented
The process leaves a trail: iteration logs, paste-sheets, session notes, generation histories, and stopping rules. Selected records are shown here; fuller documentation is available for review via [email protected].
v1-v2: first-draft lyric, then a revision pass; the shrinking outro is replaced with a SPILL outro that rises.
v3-v5: Style field compacted; the trailing NOT-list moves into a dedicated Excludes field.
v6-v7: inline negatives dropped for positive saturation; jargon audited down to plain English.
v8 LOCKED: folktronica moved to the front of the prompt. Four v8 gens ran sick in a row; Style consistency confirmed, harvest across all four.
v9 abandoned: pushed for one more refinement after v8 already worked. Paste-sheet rolled back from v9 to v8. Filed as the STOPPING RULE: when a version lands across three or more gens, stop iterating and move to production.