Describe the sound, then hear it

The generation type dropdown. It decides which models the picker below will offer.

The music workspace opens on a dropdown with two entries, Text to Music and Lyric to Music. They are not two routes to the same place. Each carries its own model roster, and the form beneath rebuilds when you switch.

Text to Music is the one that writes both the words and the music for you. If you already have lyrics and want them sung as written, the other entry is the one you want.

Text to Music runs a single model, Suno Music, and the label beside it states what a run gives you back. Two tracks, style guided.

Two tracks per generation is the detail worth planning around. Every run returns a pair of variations built from the same settings, so comparison is built into the workflow rather than something you pay for twice.

One model on this generation type, labelled with what it returns.

Seven versions. v5.5 is the default and the one most work should use.

A second dropdown holds seven versions, running v5.5, v5, v4.5 All, v4.5 Plus, v4.5, v4 and v3.5. This control is separate from the model picker and easy to miss.

v5.5 is the default. The older versions stay available because their sonic character differs, which is useful when you want a retro or lo fi quality that the current version cleans away.

Style is the only required field on this form and it holds 120 characters, which is deliberately short. It wants a list of descriptors rather than a sentence.

The buttons underneath append a tag to whatever is already there, covering cinematic, orchestral, epic, emotional, ambient, electronic, rock, pop, jazz, classical, hip hop, EDM, country, folk, metal and punk. Combining two or three buttons with your own instrument or tempo terms gives tighter control than either method on its own.

The Style field, with tag buttons that append rather than replace.

One field, two jobs, decided by the Instrumental switch further down.

This field runs to 4,100 characters and changes meaning depending on a switch lower in the form. With Instrumental off, whatever you write here is treated as lyrics and sung.

With Instrumental on, it reads as a thematic description instead. For purely instrumental work the Style field still carries more weight, so this one is best used for the mood or narrative the piece should follow.

Negative Tags works the way a negative prompt does on an image model. It names the qualities you do not want to hear, and it arrives already carrying distortion and low quality.

Exclusions belong here rather than in the Style field. Asking the main description to avoid something tends to introduce it, while listing it as a negative tag works as intended.

Negative Tags arrives pre-filled rather than empty.

Title and Persona ID, both revealed by Custom Mode.

Title takes 200 characters and does more than label the file. The name frames how the model approaches the composition, so a descriptive title is worth writing rather than leaving blank.

Persona ID offers Style or Voice, and it applies to v5.5 through v5 only. Both fields appear only when Custom Mode is switched on.

Style Weight runs 0 to 1 and starts at 0.65. High values hold the output tightly inside the genre and texture you described. Low values give the model room to interpret.

When a generation keeps drifting away from the brief, this is the first slider to raise. When every run sounds like the same safe idea, lower it.

Style Weight. All three sliders share the same range and default.

Weirdness. Higher is not better, it is stranger.

Weirdness also runs 0 to 1 from a 0.65 default. Below 0.5 you get familiar chord movement and predictable arrangement. Above 0.8 the model reaches for unusual intervals and odd time signatures.

For anything sitting under dialogue, keep it between 0.5 and 0.65. A raised Weirdness value is a creative choice rather than a quality improvement, and it makes a cue harder to cut against picture.

Audio Weight is the last of the three sliders, on the same 0 to 1 scale and the same 0.65 default. Higher values favour a clean, defined mix. Lower values allow a rawer character.

Leaving all three at 0.65 is a reasonable starting position. Moving one at a time is the only way to hear what each is actually doing, because their effects overlap.

Audio Weight, the third of the three, trading polish against variety.

Two switches. Each one changes which other controls exist.

Custom Mode unlocks the Title and Persona ID fields. With it off, those controls are not on the form at all, which is why they can appear to be missing.

Instrumental removes the vocal layer entirely, and it is the correct setting for underscore and background beds where a singer would fight the dialogue. Switching it off reveals a Vocal Gender selector offering Male or Female.