Metal AI
Music

This publication does not exist to celebrate a finished album. It exists to document a process — specifically, the process of building something complete under conditions that discourage completion.

The project at the centre of this record is a nine-track metal album produced almost entirely using AI-assisted tools, running on a mid-range Android device, by a single independent creator operating without studio access, professional equipment, or institutional support. The album was conceived, composed, arranged, visually branded, and distributed across a span of weeks.

What follows is an honest reconstruction of that process. Not a tutorial. Not a manifesto. A record of decisions, reversals, failures, and forward motion — the kind of documentation that is only possible once the dust has settled and the work has shipped.

The intended audience is anyone attempting creative work at the intersection of technology and art — designers who build things alone, musicians who code, engineers who make, students who have not yet been told what they cannot do. The hope is that someone reads this six months from now, or six years, and finds the underlying logic still useful even if the specific tools have changed entirely.

The central argument is stated plainly: artificial intelligence should amplify human creativity — not replace it. Every decision in this project was made by a human being. The AI was a collaborator, a mirror, and an instrument. It was never the author.

The question that started this project was not a technical one. It was an aesthetic one: can heavy music — music built on weight, aggression, and emotional extremity — survive contact with artificial intelligence without becoming hollow?

Metal, as a genre, has always defined itself partly through authenticity. The distortion is real. The anger is real. The physical performance — the drumming, the screaming, the precise guitar technique — is earned through years of bodily practice. The concern, legitimate and widely shared, is that algorithmically generated music strips away everything that gives the form its meaning.

This project began as an experiment to test that concern directly. Not to refute it, but to understand it through practice. The hypothesis was not that AI could replace the soul of metal. The hypothesis was that a human creator, using AI as a tool, could produce something with genuine aesthetic intent — something that could stand beside, if not inside, the tradition.

The scope evolved considerably over the course of production. What began as a loose collection of sonic experiments became a structured conceptual project with recurring themes, a visual language, and a deliberate publication strategy. That evolution — from experiment to artifact — is itself one of the central subjects of this record.

The pairing of artificial intelligence and metal music is, on its surface, absurd. Metal emerged from working-class frustration, from electric guitars played through broken amplifiers, from voices pushed past their physical limits. Its aesthetic is built on texture that cannot be synthesised: tube distortion, the resonance of a drum hit in a particular room, the biological imprecision of a human vocal cord at maximum exertion.

And yet the combination is generative precisely because of the tension. The encounter between a tool that processes pattern and a form that resists pattern-reduction produces friction. That friction is where interesting things happen.

What AI brings to heavy music

The practical contributions are concrete. AI-assisted tools can generate compositional scaffolding at a speed no individual composer can match — chord structures, rhythmic frameworks, harmonic ideas that would take weeks to develop through conventional practice. They can model existing genre conventions with enough fidelity to be useful as starting points. They can generate lyric drafts that capture thematic territory without being beholden to a single writer's vocabulary.

More importantly for an independent creator working without collaborators: they can function as an intelligent sounding board. The absence of a band, a producer, or a mixing engineer is one of the defining constraints of independent music production. AI tools partially fill that gap — not by replacing the expertise those roles represent, but by offering responsive feedback at any hour, without fatigue, and without interpersonal friction.

What metal brings to AI creative practice

Metal's extreme aesthetic demands force discipline on the AI workflow. A tool that generates competent-but-generic output is insufficient — generic metal sounds like nothing. The genre's conventions are specific enough that failures are immediately audible. This makes it an unusually honest test environment for AI-assisted production: you cannot hide behind pleasant adequacy.

The combination also forces a clear delineation of what the human creator contributes and what the tool contributes — a delineation that is useful to document, because it maps onto questions that are relevant far beyond this project.

Before a single piece of audio was generated, the project required a research phase that was more curatorial than technical. The question was not which tools existed — that landscape is well-documented and changes rapidly — but which tools were appropriate for the specific constraints of this project.

The constraints were significant. The production environment was a single mid-range Android device. No desktop DAW. No cloud rendering. No professional plugins. The tools had to be lightweight, mobile-compatible, and either free or low-cost. The workflow had to be manageable by one person without prior professional music production training.

What the research revealed

The most significant finding was not about tools. It was about the existing body of AI-assisted music production. Much of it clusters in two categories: ambient and electronic music, where the algorithmic origin is aesthetically congruent; and novelty content, where the AI origin is the point. Very little serious independent work had been done at the intersection of AI assistance and heavy guitar music.

That gap was both a challenge and an opportunity. There were few reference points for what success looked like. But there were also few expectations to manage and few conventions to violate.

The thematic framework

The research phase also established the conceptual architecture of the album. The dark-fantasy thematic territory was chosen deliberately: it provided narrative coherence across tracks, gave the lyrical AI tools sufficient context to generate material with internal consistency, and aligned with metal's established relationship with mythology, mortality, and transformation.

Track one, "To Face You Death," was identified early as the project's anchor — the piece that would establish tone, visual language, and production standard for everything that followed.

The first serious failure of the project was the absence of a workflow. The initial impulse was to begin generating immediately — to treat the tools as infinitely available and the output as automatically coherent. That impulse produced chaos within hours.

The outputs were disconnected. Tracks had no relationship to each other. Lyrical themes contradicted sonic character. Visual ideas were generated before the audio had any form. The project was producing material without producing an album.

The three-mode operational matrix

The resolution was structural. Three distinct operating modes were defined and enforced as hard rules — not guidelines, but states that could not be blurred without cost.

The mode separation was the single most important structural decision in the project. Without it, every session collapsed into a recursive refinement loop that produced nothing deployable. With it, each session had a defined objective and a defined exit condition.

The over-optimization problem

The workflow design also addressed a specific, recurring defect: the tendency to extend refinement cycles past the point of functional improvement. The solution was a hard iteration limit: seven cycles for logic configurations, three cycles for visual layouts. When the limit was reached, the current version was frozen and the session moved to Execution Mode. Done became the baseline requirement, not the fallback.

Prompt engineering is a misleading term for what the practice actually involves. It suggests a technical skill — a set of formulaic inputs that produce reliable outputs. The reality is closer to directed conversation: the ability to establish context, constrain scope, and calibrate specificity in a way that guides a generative system toward useful territory without over-determining the result.

Phase One — Abstract Input

Early prompts were conceptually rich but structurally vague. Thematic territory was described in broad terms: darkness, transformation, confrontation, mythological scale. The outputs were correspondingly broad — technically competent but undifferentiated. They could have been for any heavy music project, not this one.

Phase Two — Structural Specification

The shift to structural specification changed the output quality substantially. Prompts began to include explicit parameters: tempo range, key signature, subgenre markers, specific instrumentation, lyrical meter, thematic anchors from earlier tracks. The outputs became more coherent and more specific to the project's evolving identity.

Phase Three — Iterative Dialogue

The most productive phase was not the most complex prompt. It was the dialogue — the process of using initial outputs as context for subsequent prompts, building a feedback loop that accumulated specificity over multiple exchanges. The album's thematic vocabulary, visual language, and sonic character emerged from this iterative dialogue rather than from any single prompt.

Design principles applied to prompting

Lyric generation was the most contentious part of the AI-assisted workflow — the area where the tension between human intent and machine output was sharpest, and where the ethical questions were most immediate.

The starting position was clear: AI-generated lyrics would function as drafts, not finals. The voice of the album was a human voice. The machine's contribution was material — raw vocabulary, structural scaffolding, thematic suggestions — that the human creator then shaped, revised, and claimed.

The dark-fantasy vocabulary

The thematic architecture of the album — confrontation with mortality, mythological transformation, the psychology of extremity — gave the lyric generation tools enough context to produce material with internal consistency. Track one established the central image: direct confrontation with death as a force rather than an event. "To Face You Death" set a tone of agency — the subject is not fleeing or mourning, but standing in confrontation. That posture recurred across the album's thematic arc.

Meter, cadence, and the metal vocal

A specific constraint shaped all lyric development: the output had to work as metal vocal performance, which means it had to fit the rhythmic structure of the composition and be physically singable at the intensity the genre demands. This constraint eliminated a significant proportion of AI-generated material — lyrically interesting text that would be metrically or physically unsuitable for performance.

The discipline of testing every lyric draft against the compositional rhythm before finalising it was essential and time-consuming. The judgment that something sounds right in the mouth, rather than right on the page, is not something the tools could provide.

Composition in this context refers to the generation and selection of the core musical material — the harmonic structures, melodic ideas, and rhythmic frameworks that define each track before arrangement or production decisions are applied.

The compositional tools available on a mobile device are narrower than their desktop equivalents, but this constraint proved generative. Limitations on polyphony, on instrument layering, and on processing overhead forced compositional simplicity — and simplicity, in heavy music, is often a virtue. The most effective metal riffs are not complex. They are precise.

Working in the dark-fantasy idiom

The compositional territory of dark-fantasy metal draws on a specific vocabulary: minor and diminished harmonies, chromatic movement, modal interchange, drop-tuning that emphasises the lower register. These are learnable conventions. The AI tools could model them with reasonable fidelity.

What the tools could not model was the compositional character that distinguishes a track from a template. That required editorial judgment — the ability to listen to generated material and identify not the best-executed output, but the output with the most potential to develop into something with genuine character.

The nine-track arc

The compositional challenge of a full album is different from the challenge of a single track. Individual tracks need internal coherence. An album needs an arc — a sense of movement from opening to close that gives the collection a reason to exist as a collection rather than a playlist. That arc was designed before composition began and revised as composition produced material that suggested different directions.

Arrangement is the discipline of deciding what happens when. Which instruments enter and exit, where the dynamics build and fall, how the verse-chorus-bridge structure serves or departs from the genre's established conventions. It is, in many ways, the most compositionally sophisticated part of production — and the area where AI assistance was least reliable.

Generated arrangements tended toward predictability. The tools modelled genre conventions accurately enough to produce structurally correct outputs — verse, pre-chorus, chorus, breakdown, bridge — but the accuracy was the limitation. Metal's most memorable moments tend to occur where the expected structure is interrupted: an unexpected metric shift, a dynamic collapse that violates the build, a section that refuses resolution.

Human intervention as structural design

The human contribution to arrangement was primarily subtractive. Generated structures were simplified, sections were removed, and conventional resolutions were replaced with more ambiguous or aggressive alternatives. The AI provided the skeleton. The human creator broke bones to reshape it.

Sound design in heavy music refers to the tonal character of individual elements: the specific quality of distortion on the guitar, the character of the drum sound, the processing applied to the vocal, the ambient texture that sits beneath or between the primary elements. These decisions are, collectively, what makes a metal recording sound like itself rather than like any metal recording.

Mobile production constraints as aesthetic parameters

Audio produced on consumer mobile hardware has a characteristic sonic signature: a compressed mid-range, limited low-frequency extension, a brightness that comes from smaller drivers and digital processing chains. Rather than attempting to overcome this signature, the project incorporated it. The resulting sound is not what a professional studio would produce. It is something else — a record of what is possible with the tools that are available.

The dark-fantasy atmosphere

The conceptual territory of the album demanded specific atmospheric qualities: a sense of physical weight, of temporal scale, of something ancient and threatening. These qualities were pursued through layering — the accumulation of low-level textural elements beneath the primary audio that created an impression of depth without requiring the processing headroom of complex reverb chains.

For track one, the approach was a frozen keyframe with a rain-particle overlay — a visual and sonic simplicity arrived at through the failure of more complex approaches. The simplification reduced export time by approximately 70% while preserving the atmospheric character the track required. This was not a compromise. It was an editorial decision.

The most honest chapter of any creative process documentation is the one that describes what failed. Iteration is not refinement toward a predetermined ideal — it is a process of discovering what the work is, through the accumulation of attempts that are insufficient.

The concurrent rendering failure

The most significant technical failure in the project occurred during the audio-visual production phase for the full nine-track package. The initial plan was concurrent rendering — all nine tracks processed simultaneously to produce a unified output. On the Redmi 12, this triggered thermal throttling within minutes. The device's memory allocation was saturated. The rendering process terminated.

The resolution was structural rather than technical. Concurrent rendering was abandoned. A sequential deployment protocol was implemented: isolate a single track, freeze its configuration, execute the export, then proceed to the next. Track one was completed within a 48-hour window. The protocol established a rhythm for the remaining eight tracks.

The discovery loop

The productive failures were not only technical. Several compositional directions were pursued across multiple sessions before being abandoned — not because they were technically failed, but because they did not serve the album's evolving identity. These sessions clarified what the album was not, which is a necessary step in understanding what it is. The documentation of abandoned directions is as important as the documentation of what succeeded.

Visual identity in music production is often treated as secondary — something applied after the audio is complete, to package it for distribution. That approach was explicitly rejected in this project. The visual language was developed in parallel with the sonic material, and the two were designed to be mutually reinforcing.

The dark-fantasy thematic territory established parameters for the visual direction as clearly as it did for the audio. The visual language needed to carry a sense of weight, antiquity, and scale — communicating the album's emotional register before a single note was heard.

Constraint-driven visual design

Complex generative visual systems — dynamic backgrounds, particle effects with high frame counts, layered compositing — were ruled out by the device's processing limitations. The visual approach had to be atmospheric through simplicity rather than through complexity.

The rain-particle overlay on track one became a visual signature: a single dynamic element applied over a high-contrast still image, creating movement and atmosphere without rendering load. It was a constraint solution that became an aesthetic choice.

The pixel-timer UI element

A stylised pixel-timer UI was incorporated into the track one visual — a deliberate anachronism that placed a digital, grid-based element within the dark-fantasy visual territory. The combination was an example of the productive friction that defined the broader project: the collision of the algorithmic and the atmospheric, the modern and the ancient.

Album artwork occupies a peculiar position in music production: it is not the music, but it is the first encounter most listeners will have with the music. It is the surface through which the work enters the world, and the image by which it will be remembered — or not.

Multiple directions were generated and evaluated before a direction was committed to. The evaluation criteria were consistent: does this image communicate the album's emotional register? Does it extend the visual language established for track one? Does it have the scale and weight the music requires?

AI-assisted image generation

AI image generation tools were used extensively in the artwork development process. The tools were productive at generating material in the thematic territory — dark landscapes, mythological figures, atmospheric imagery — with speed that would have been impossible through conventional commission or original photography.

The publishing phase produced a lesson that was, in retrospect, predictable: creative completion and public visibility are two entirely separate problems, each requiring its own discipline.

The first track was published with abstract, internally coherent branding — titles and metadata that made sense within the project's own vocabulary but communicated nothing useful to external discovery systems. The result was zero initial views. Not a failure of the music. A failure of metadata.

The algorithmic distribution overlay

Public discovery networks parse structural markers, keyword density, and engagement pathways. They do not parse artistic intent. A technically excellent piece of music with unoptimised metadata is invisible to these systems.

Three distribution channels were activated in parallel: long-form video with explicit subgenre SEO tags; 9:16 short-form extraction as a discovery vector; and direct engagement with specialised technical forums where the production methodology itself was relevant to the audience.

Translating creative output into professional capital

The final distribution challenge was institutional: how to present the project to professional networks in a way that communicated its genuine technical complexity. A music production project, presented as such to a corporate audience, reads as a hobby. The same project, presented as a case study in resource-constrained media engineering, reads as a demonstration of professional competency. The re-framing was not dishonest. It was translation.

The following lessons are stated without elaboration. Each is grounded in a specific decision or failure documented elsewhere in this record. They are offered as transferable principles rather than project-specific conclusions.

A credible case study documents its limitations as carefully as its achievements.

The question of what AI can and cannot replace in creative practice is better answered through documented practice than through theoretical argument. This project constitutes a body of evidence.

What AI replaced or significantly augmented

What AI could not replace

Future directions are stated as genuine possibilities grounded in the evidence of this project — not as aspirational claims or marketing language. Each is identified as a logical extension of what was established here, conditioned by the limitations that were also documented here.

Expanded hardware access

The most immediate and most significant constraint on this project was hardware. The sequential deployment protocol, the visual simplification decisions, and the audio fidelity limitations are all direct consequences of mobile-only production. Access to modest desktop hardware — not professional studio equipment, but a mid-range computer with a standard DAW — would substantially expand the production possibilities while leaving the core workflow largely intact.

Collaborative integration

This project was produced by a single individual. The workflow was designed for solo operation. The next iteration could incorporate human collaborators — a vocalist, a guitarist, a mixing engineer — who interact with AI-generated material at specific points in the pipeline. The documentation discipline established in this project would need to extend to capture the collaborative decisions as precisely as the solo ones.

Real-time AI integration in performance

The tools used in this project operated in a generation-and-selection mode: produce material, evaluate it, select what to keep. A different application of similar technology would integrate AI into real-time performance — using generative systems to respond to live input in ways that extend rather than replace what the performer is doing. This is a substantially different technical and artistic problem, but one adjacent to the work done here.

The documentation as a living system

This publication is designed to be updated as the project evolves. The NM Studio Documentation Library is an ongoing project, not a closed record. Future volumes will document subsequent production cycles, refined methodologies, and the outcomes of directions described here. The system is designed to be calibrated continuously — not fixed at the moment of first publication.

This publication does not close a question. It documents the first serious attempt to answer one — and the record of that attempt is itself the contribution.

The project established that AI-assisted metal production is possible. That a single individual, operating on consumer mobile hardware, can produce a nine-track conceptual album with genuine aesthetic intent and a coherent visual identity. That the tools exist, that a workflow can be designed around their constraints, and that the result can be published and distributed through conventional channels.

It also established, with equal clarity, what AI cannot do in this context. It cannot make the decisions that determine whether the work is specific rather than generic, genuine rather than merely competent. It cannot bear the creative responsibility that makes the work belong to someone. It cannot stand behind the work when it meets the world.

What remains is the next cycle. The calibration loop is not a diagram. It is a practice. Every project is an input to the next iteration of the system. This publication is both an output of the first cycle and a structural input to the second.

The work continues. The record continues. The system is designed to outlast any single version of itself.