Fine-tune your brand

Set up your brand gets you the essentials. This page is the deep dive: every field on Settings → Brand → Voice (visible once the AI Lyric Generator is enabled), what it does, and how to fill it in so the generated lyrics sound like a coherent band rather than a random prompt.

Everything on this page lives in the Voice tab. Fields auto-save as you type.

How the Voice fields work together

The LLM reads your brand as one block of creative direction every time it writes a song. Each field plays a specific role:

Field	Role
Genre	The musical lane — hip-hop, alt R&B, dark pop, hardwave, etc.
Vocal type	Who sings, and whether it’s a solo or a duet exchange.
Your Sound	The narrator’s character — perspective, attitude, lens.
Explicit content	How direct the lyrics are allowed to get.
Vocal persona (optional)	A short timbre description added to the Suno style prompt to fight voice drift.
Themes (advanced)	Emotional territories the brand rotates through.
Never use (advanced)	Words, images, or tropes to avoid.
Always prefer (advanced)	Detail types to lean into (proper nouns, times, body-level detail).
Artist references (advanced)	Sonic neighbours — inspiration, not imitation.

Think of them as filters stacked on top of each topic you give the generator. A vague brand produces vague lyrics. A specific brand produces distinctive lyrics even from a one-line topic like “the last time I saw him”.

Genre

Pick a preset lane from the dropdown (Dark Pop, Alt R&B, Hardwave, Trap, etc.), or choose Custom… and write your own short description. A custom line might read:

dark trap pop with hardwave edges and R&B vocal phrasing

Keep it to a single sentence — the LLM doesn’t need a music-theory essay, it needs a lane label. Custom is for when no preset quite captures your sound; otherwise a preset is fine and is tested against the prompt templates.

Vocal type

Five options:

• Female — single female vocalist.
• Male — single male vocalist.
• Duet (male + female) — two voices trading off, most common duet shape.
• Duet (male + male) — two male voices.
• Duet (female + female) — two female voices.

For duets, the generator writes parts that alternate lines and sometimes overlap on hooks. If you’re generating for Suno, your chosen vocal type is also included in the style prompt so the model knows who’s singing.

Your Sound

The most important field on the page. Two to four sentences is plenty — no word limit enforced, but longer isn’t better. You’re describing the narrator, not the music.

Hit three beats:

1. Perspective — who they are. Role, stance, situation.
2. Attitude — what they feel toward their subject. Amused, bitter, possessive, tender, dismissive.
3. Lens — what they notice that others wouldn’t. Specific details they fixate on.

Examples

Too vague:

Warm, nostalgic, thoughtful.

No narrator, just weather. The LLM has nothing to write toward.

Good — small-town observer:

A songwriter from a town everyone else left. Stayed on purpose, not by default. Writes about the diner regulars, the seasons, and the cars that pass through without stopping. Prefers specifics over metaphors — a street name, the smell of rain on hot asphalt, the exact hour the light changes on a Sunday afternoon.

Clear perspective (rooted local), clear attitude (chose this, not stuck with it), clear lens (concrete detail over metaphor).

Good — returning traveller:

A former tour roadie who left the industry and now writes songs about the people who stayed. Warm but unsentimental. Notices hotel-carpet patterns, the way a tour bus smells on day nine, the difference between a handshake from someone who’s made it and someone who hasn’t.

Specific job, specific stance, specific sensory fingerprint.

Good — second-chorus lift:

A narrator who keeps things quiet through the first half of a song and lets the feeling build to the second chorus. Prefers specifics over metaphors — a porch light, a train schedule, a coffee gone cold on the counter. Never uses the word “soul.” Writes about people who show up.

Describes the structure of how the narrator emotes, which the LLM can pick up and mirror in the arc of the lyric.

Prompt checklist

Before saving, skim your Your Sound and ask:

• Could this describe ten different artists? If yes, push for more specifics.
• Does it mention a situation (a job, a relationship, a stance)? Abstract moods don’t write songs — people in situations do.
• Does it say what the narrator won’t do? Constraints are as useful as permissions — “never uses the word soul” is gold.

Explicit content

A single dropdown with three values. This is a policy decision, not a vibe; pick once and let every song follow it.

• Off — The LLM never produces profanity or overt sexual language, even if the topic invites it. Mature topics are softened or sidestepped — a breakup revenge song reads more as frustration than rage. Choose this for family-friendly channels or if your platform distribution requires clean content.
• Contextual (recommended default) — Let the topic decide. Heavy or mature topics can earn one or two strong words where it serves the scene; light topics stay clean. Feels natural rather than forced in either direction. Best setting for most artists.
• On — Allow direct language freely when it fits. The LLM won’t soften profanity, sexual tension, or aggression. Best paired with the Uncensored model tier (Cydonia 22B v2q) — aligned models like Qwen or Mistral will still self-censor regardless of this setting, so leaving this on while running the Balanced model will produce the same output as Contextual in practice.

If you need explicit output reliably, you need two things: this set to On and the Uncensored model selected in Settings → Lyrics.

Vocal persona

Below the core fields is a section about vocal persona — this exists to help you fight voice drift in Suno (or similar tools) between generations.

The problem: Suno can assign a different voice on each generation, even with the same style prompt. If you’re trying to build a catalogue that sounds like one artist, that’s a problem.

Two levers the app gives you:

Lever 1: Include a voice description in the style prompt

Toggle Include persona description in the style prompt on, then fill in Voice description. The text you write gets appended to every song’s Suno style prompt. This doesn’t eliminate drift but it cuts it significantly.

Specific timbre words beat vague ones. Try:

• Register — alto, soprano, baritone, tenor.
• Texture — raspy, breathy, smoky, bright, gravelly.
• Delivery — conversational, belting, restrained, half-whispered, rhythmic.

Short is fine — one comma-separated line:

low breathy alto, smoky delivery, restrained dynamics, conversational half-whisper

Lever 2: Lock a Suno Persona (the real fix)

Once you have a batch of Suno generations, pick the one whose voice most sounds like “the brand” and turn it into a Suno Persona. The in-app settings pane has an expandable “How to lock a Persona in Suno” section; the short version:

1. In Suno’s web UI, open the winning song, click the ⋮ menu, choose Make Persona.
2. Name it after your brand (e.g. “Nightshade v1”) and save.
3. On future generations, select that Persona from Suno’s dropdown. The voice stays consistent.
4. Optional: once the Persona is reliable, you can turn the in-app persona toggle back off — the Persona itself carries the voice, and the description becomes redundant.

The app includes this workflow because until the Suno integration ships, you’re still pasting into Suno manually and need a way to keep voices in line.

Advanced — the creative fingerprint

Scroll down and click Show advanced fields. Four list-style textareas, one item per line. Optional, but filling these in is where a brand stops sounding like “a singer with good Your Sound” and starts sounding like your singer.

Themes you gravitate toward

Emotional territories the brand rotates through. The generator uses these to keep songs from landing on the same emotional beat every time — if “returning to a place that’s changed” is one of your themes, it’ll surface in ~1 in 3 songs rather than ~1 in 10.

returning to a place that's changed since you left
learning something late from an older relative
the small daily rituals that make a place feel like home
friendship that outlasts distance
persistence through a slow season

Never use

Words, images, or tropes this brand avoids. A constraint list. As important as themes — tells the LLM the negative space.

heart-on-my-sleeve cliches
weather as a stand-in for emotion
"your eyes" as the central image
the word "soul"
anything that sounds like it's from an ad

Always prefer

Detail types you want the lyrics to lean on. This shapes the texture of every line, not the subject. Proper nouns, exact times, body-level specificity — these are the fingerprints that make lyrics feel lived-in rather than written.

proper nouns (names, towns, streets)
specific times of day and seasons
one-line observations instead of extended metaphors
concrete objects the narrator can actually touch

Artist references

Artists or songs you want to sound adjacent to. Inspiration, not imitation — the LLM uses these as a sonic neighbourhood, not as targets to copy.

Noah Kahan — Stick Season
Gregory Alan Isakov
The Head and the Heart

Don’t stack twenty references; three to five is plenty. Too many references muddy the signal.

Putting it all together

A filled-in brand that sings:

• Genre: Indie folk with singer-songwriter warmth
• Vocal type: Male
• Your Sound: A songwriter from a town everyone else left. Stayed on purpose, not by default. Writes about the diner regulars, the seasons, and the cars that pass through without stopping. Prefers specifics over metaphors.
• Explicit: Off
• Vocal persona: warm baritone, unpolished edges, conversational delivery, gentle vibrato on held notes
• Themes: returning to a place that’s changed / learning something late from an older relative / the small daily rituals that make a place feel like home
• Never use: heart-on-my-sleeve cliches / “your eyes” as the central image / weather as a stand-in for emotion
• Always prefer: proper nouns (towns, streets) / specific times of day and seasons / concrete objects the narrator can actually touch
• Artist references: Noah Kahan — Stick Season / Gregory Alan Isakov / The Head and the Heart

Feed that brand a one-line topic like “the first snowfall after I moved back” and the generator has a person to write as, a lane to stay in, and constraints to bump against. That’s what produces lyrics that don’t sound like anyone else’s.

Editing or replacing the brand

Everything auto-saves, and you can overwrite anything at any time. Songs you’ve already generated keep their original brand baked into their style prompt and chosen lyrics — editing the brand only affects future generations. If you’re experimenting, you can also treat a single brand as your “main lane” and swap in temporary overrides for one-off projects rather than trying to keep one brand that covers every mood.

Next steps

• AI Lyric Generator — the full feature reference, including the scene-then-song generation flow.
• Make your first video with AI lyrics — end-to-end walkthrough once the brand is set.