AI-Assisted Music Production: The Saint-Kitsch Method

Most conversations about AI-assisted music production start with the tool and end with the output. Ours starts with a person banging on a hardware synth in a kitchen, and ends with a record someone chose to keep.

Saint-Kitsch is one of the artists on Prompting Records, a label built on the premise that AI-native music is a production condition, not a genre. The Saint-Kitsch method is the most documented production system we run. This article describes it in full — not as a tutorial, but as a working definition of what ai-assisted music production looks like when a human stays accountable for every decision that matters.

What ai-assisted music production means here

The term gets used two ways. In the weak sense, it means a person typed a prompt and a model returned a song. In the strong sense — the one we defend — it means a production pipeline where generative tools occupy specific, bounded stages, and human judgment occupies the rest.

The difference is structural, not moral. A pipeline that delegates everything to a model has no author. A pipeline that delegates finishing, separation, or texture to a model — while a human owns the material, the structure, the selection, and the final mix — has an author who happens to work faster.

Saint-Kitsch operates entirely in the strong sense. Deadpan punk techno, 148 BPM, French spoken vocals. Every track that ships went through five stages, and only one of them belongs to a generative model.

The method, stage by stage

The first stage is raw material, and it is fully human. Saint-Kitsch starts from hardware synth improvisation and spontaneous singing — first takes, captured as they happen, in the middle of ordinary life. Nothing is written for the machine. The material exists before any model sees it. This is the part of the chain we treat as sanctuary: the source is a person, recorded in the act, not a prompt describing a person.

The second stage is the daily journal. Saint-Kitsch is built as Everyday Music — a musical diary rather than a polished product line. Fragments of daily life, filmed and recorded, feed the writing. The lyrics follow a fixed skeleton: spoken deadpan verses, a punk chorus, a bridge where a female voice answers the monotone, an outro that dissolves into melancholy. The structure is non-negotiable. Constraint is what makes a journal legible as a body of work instead of a pile of takes.

The third stage is where AI enters, and where it stops. Generative audio models are used for finishing — turning raw fragments and structural intent into full arrangements. The model proposes; it never decides. Every generation is disposable by default. Most are discarded. The ones that survive do so because they passed a taste decision, not a quality threshold a machine could evaluate.

The fourth stage is reconstruction. Surviving generations are separated into stems and rebuilt inside a conventional DAW session. Vocals are treated, resonances are tamed, saturation is added by hand, the bus is compressed the way engineers have compressed buses for forty years. This stage matters because it reverses the relationship: the AI output becomes raw material again, and a human engineer makes it a record. The model's voice is in there, but the fingerprints on the final master are ours.

The fifth stage is curation, and it is the hardest to automate away. Every Saint-Kitsch ship passes a brand coherence review against the artist's documented identity — six fixed aesthetic pillars, a locked visual palette, a recurring character, two canonical symbols. A track that sounds right but belongs to nothing gets cut. We wrote elsewhere that taste cannot be prompted; this stage is that sentence turned into a gate.

A journal, not a product line

The phrase that governs the whole system is Everyday Music. Saint-Kitsch does not produce singles the way a content calendar produces posts. It keeps a diary, and some pages of the diary become records.

This matters for ai-assisted music production because the dominant failure mode of generative tooling is volume. A model can produce three hundred plausible tracks in an afternoon. A label that ships them has confused capacity with catalogue. The journal frame inverts the incentive: the question is never how much the system can generate, but which fragment of an actual life deserves finishing. Output drops. Identity compounds.

The earlier Saint-Kitsch methodology note sketched this position when the project began. What has changed since is that the method is now fully documented internally — source capture, writing formula, finishing boundaries, mix chain, coherence review — to the point where any engineer on the label could execute it without diluting the artist. That documentation is itself part of the method. A production system that lives in one person's head is a habit. A production system written down is a label asset.

Why the human stays in the chain

There is a practical argument and a positional one.

The practical argument: generative audio is good at texture and bad at intent. It renders a style; it does not hold a worldview. Saint-Kitsch's worldview — absurdist, deadpan, kitsch on purpose — lives in decisions a model has no access to: which first take is funny, which lyric is too explanatory, which chorus betrays the character. Removing the human does not remove the decisions. It just makes them randomly.

The positional argument: everything Prompting Records releases is creatively coded by humans. The talent is human curation. That is the line that separates a label from a feed. AI-assisted music production, done in the strong sense, is not a compromise between human craft and machine speed — it is a reallocation. The machine absorbs the labor. The human keeps the authorship.

What the method refuses

The method is defined as much by exclusions as by stages. It refuses full delegation — no track ships that was generated end-to-end without human source material. It refuses volume as strategy — the catalogue grows by selection, not by accumulation. It refuses style drift — the writing formula and the aesthetic pillars are fixed, and deviation requires an explicit, documented decision. And it refuses the demo aesthetic — AI output is never shipped raw, because raw model output sounds like the model, and the only acceptable sound is the artist.

These refusals are what make the method portable. Strip the Saint-Kitsch specifics and the skeleton remains: human source, constrained writing, bounded generation, human reconstruction, identity gate. Any artist on the label can instantiate it with different aesthetics. The next one already is.

FAQ

What is ai-assisted music production?

AI-assisted music production is a workflow where generative models handle bounded stages of making music — typically arrangement, finishing, or sound design — while humans own the source material, structure, mixing decisions, and final selection. It is distinct from fully generative music, where a model produces the complete work from a text prompt.

How is this different from AI-generated music?

AI-generated music starts and ends with the model. AI-assisted music production starts with human material — recorded performance, written structure, documented intent — uses a model as one stage among several, then returns the output to human hands for reconstruction and curation. The author is a person; the model is a tool with a fence around it.

Does AI replace the producer in this workflow?

No. In the Saint-Kitsch method, the producer's role shifts rather than disappears: less time rendering, more time selecting, structuring, and mixing. The decisions that define a record — what to keep, what it should belong to, when it is finished — remain human at every stage where they occur.

Can this method scale to other artists?

Yes, by design. The method's skeleton — human source, constrained writing, bounded generation, human reconstruction, identity gate — is artist-agnostic. Each artist instantiates it with their own aesthetic doctrine. The constraint that does not scale is taste, which is precisely why it stays human.

The machine absorbs the labor. The author keeps the name.