Style Influence and Weirdness: the two sliders that decide your result

Two people write the same Style prompt and get results that sound nothing alike. Usually the difference is not the prompt at all. It is a pair of sliders sitting in the advanced options of custom mode, which most people never touch and which quietly decide how literally the model reads what you wrote.

Style Influence controls obedience. Weirdness controls risk. Audio Influence, when you have given the model a reference, controls how tightly it holds to the source. Learning what each one changes is worth more than another hour of adjective shopping.

What each slider actually does

Style Influence: how literally your prompt is followed.

Low, and the model treats your Style string as a suggestion. It reaches for what usually works in that genre, which is another way of saying it reaches for the average. Your carefully specified dusty upright piano becomes a piano.

High, and the model tries to satisfy every token you wrote. Every instrument, every production adjective, every era marker. This is what you want after you have written a precise prompt, and exactly what hurts when your prompt contains contradictions, because now the model has to honour both halves of the contradiction at once.

Weirdness: how far from the safe centre the model will wander.

Low, and you get the competent, expected version of the genre. Radio-shaped, predictable, and often the right answer for background music, a podcast bed or anything where the song is not the point.

High, and the model takes structural and timbral risks: unusual transitions, odd textures, arrangements that do not resolve where you expect. Push it far and you get granular, glitchy artefacts that either sound like design or sound like a fault, depending on the track.

Audio Influence: how closely a reference is held.

Only in play when you have supplied audio. Low lets the model drift away from the source into something new; high keeps it tethered. When an uploaded clip is the seed for the track, this is the dial that decides whether you get a variation or a departure.

The interaction nobody explains

The sliders are usually described one at a time, which hides the only thing that matters: they fight each other.

High Style Influence plus high Weirdness is a contradiction. You are asking the model to obey your prompt precisely and to take large risks at the same time. What comes back is usually incoherent, because the two instructions pull the arrangement in opposite directions.

Low Style Influence plus low Weirdness is the safest and dullest combination available. Nothing is wrong with the output. Nothing is memorable either, because you have asked for the statistical middle of a genre and received exactly that.

The two combinations worth knowing:

High Style Influence, low-to-moderate Weirdness. The working default for anyone with a detailed prompt. You wrote a specific recipe, you want it followed, and you do not want the model improvising over the top of it.

Moderate Style Influence, high Weirdness. The exploration setting. Loose brief, let the model surprise you, keep the takes that are interesting and reverse-engineer a prompt from them afterwards. This is how you find combinations you would never have written down.

Test it properly, because guessing costs credits

One-variable testing sounds obvious and almost nobody does it.

1. Write one prompt and lock it. Do not touch a word.
2. Generate at low Style Influence.
3. Generate again with only Style Influence raised.
4. Listen for one thing specifically: did the instrument you named actually appear?

That single question tells you more than a general impression of quality. If your named instrument shows up only at the high setting, your prompt is fine and the slider was the problem all along. If it never shows up, the prompt is the problem and no slider will rescue it. That is when the prompting guide earns its time.

Then repeat the same procedure for Weirdness, changing nothing else. Four generations, forty credits, and you stop guessing forever.

Keep the results. A note that reads "high style, low weird, this Style string, sounded right" is the beginning of a personal preset library, and it is the difference between a hobby and a workflow.

When a slider is the wrong tool

Three problems get blamed on sliders and are actually something else.

The prompt contradicts itself. lo-fi bedroom recording, pristine studio production cannot both be satisfied. Raising Style Influence makes this worse rather than better, because now both halves are being enforced. Fix the words. Suno ignoring your prompt is usually a conflict, not a deafness.

The prompt is a pile of adjectives. epic, emotional, powerful, beautiful gives the model nothing to hold. Adjectives describe your reaction to music, not the music. Instruments, decade, production character and vocal treatment are things a model can act on. Our catalogue is built entirely from that second kind of vocabulary.

You wanted something removed. Sliders shape emphasis, they do not exclude. Getting rid of an instrument or a texture is a separate mechanism, and negative prompting has its own rules.

Practical starting points

Not laws, just sane places to begin before you run your own tests.

• Background and functional music: low Weirdness. Predictability is the feature. A study or focus track should not surprise anyone.
• A detailed genre recipe you trust: high Style Influence. You did the work in the prompt, make the model use it.
• Hunting for an idea: high Weirdness, moderate Style Influence, and generate more takes than usual.
• Extending an existing track: keep settings consistent with the original generation, or the new section will not feel like the same song.
• Working from an uploaded reference: start Audio Influence high and lower it until the result stops sounding like a copy and starts sounding like a response.

FAQ

Where are these sliders? In the advanced options of custom mode, below the main fields. If you only use the simple description box, you never see them.

What is a safe default? Style Influence high when your prompt is specific, Weirdness low to moderate. Adjust from there with one-variable tests.

Does high Weirdness make songs more creative? It makes them less predictable. Creative is a judgment you make afterwards, and past a certain point the output is just noisy.

Do the sliders change the model? No. Model version is a separate choice, and it changes far more than any slider. Sliders steer how the chosen model interprets you.

Why do my results still vary with identical settings? Generation is not deterministic. Consistency across takes comes from structure and persona work, which is a repeatability problem, not a slider one.

Should I move sliders or rewrite the prompt? Rewrite first. A slider adjusts how a prompt is read. It cannot supply information the prompt never contained.