style-fieldsuno

The Exclude field, and what it ignores

Silas Moser5 min read
Open Studio

You typed "no autotune" and the vocal came back drenched in it. You asked for no guitar solos and the track opens with one.

The Exclude field is the most disputed control in the interface, and the dispute is worth reading before you decide whether to use it.

There is a field for this, and it is not the Style box

Exclusions have their own input. Songsmith's negative prompt guide locates it: in Custom Mode you open Advanced Options and the Exclude field sits at the top of the panel. What goes in it is instruments, styles and vocal types.

Which means the popular habit of writing exclusions into the Style prompt is doing something different from what people think. On a widely shared style guide that recommends an "Exclude:" line inside the main prompt, a reply asks the obvious question: why put exclusions in the middle of the main prompt, where they can only confuse the model, when the interface has a field for exactly that?

Nobody has measured the difference. But one of those is a field the interface reads as exclusions and the other is more words in the description of your song.

Suno's team says use it. Users say it ignores them

From the Suno team's own post for the v5 model: "Refine with exclusions: Use exclude tags to push away areas you don't want."

That post was written for v5, which Suno retired on 9 September 2026. On v6, one user reports exclusions being ignored outright, which is one more reason to retest the field on the new model before you rely on it.

From the reply section of that same style guide: for one writer it ignores nearly everything, and the exclusions most of all. A year later somebody lists what they had excluded, pinch harmonics, guitar solos, guitar fills and squeals, and asks what the first noise in the generation turned out to be.

Both can be true of a control that nudges rather than forbids. Songsmith says as much in the same breath as recommending it: "Negatives are guidance, not hard bans. Excluded elements may still appear occasionally."

Naming what you do not want puts the word in the room

The oldest complaint about this is the sharpest, and it comes from a 2024 tips thread: asking for no autotune is effectively asking for autotune.

Whether that is what happens inside the model is not something anyone outside Suno can say. What is easy to check is the writing habit it produces. Here is a Style field built out of refusals:

indie rock, no autotune, not too fast, no synths, no female vocals,
nothing electronic, moody but not sad, no pop production

Eight descriptors and five of them are negatives, which leaves the model almost nothing to aim at. Now the same intent, with the exclusions moved to the field that takes them:

Style:   indie rock, moody, electric guitar, male vocals, 100 bpm
Exclude: synths, autotune

The positives say what the song is. The two exclusions are the ones that are genuinely about absence, and everything else that was phrased as a refusal became a decision.

Style field, five of eight descriptors phrased as refusalsStyle field, all positiveindie rockno autotunenot too fastno synthsno female vocalsnothing electronicmoody but not sadno pop productionindie rockmoodyelectric guitarmale vocals100 bpmExclude fieldsynthsautotune
Tip

If you can rewrite an exclusion as something you want, do that instead. "No female vocals" is a weaker instruction than "male vocals", and it costs you a descriptor slot to say it backwards.

A v6 writer put a mechanism under that habit in September 2026, two days after the model shipped. Their reading is that the refusal goes into the conditioning text as text, so the model keeps the noun and drops the negation, which is why every instance of reverb, echo and room in a "no reverb" style prompt is still a request for reverb. It is a community reading rather than anything Suno has published, and the writing it produces is the same either way: move the refusals into the field built for them, and say the positive version in the Style box, "dry, close-mic'd", "narrow stereo image".

Two or three, and only when the positive is already right

Songsmith's rule is a ceiling of two or three exclusions, on the grounds that more than that confuse the model rather than constrain it. That matches how the other Style-field advice behaves, where more descriptors start contradicting each other rather than adding detail.

The order of work follows from that. Get the positive description right first, generate, and only then add an exclusion for the specific thing that keeps arriving uninvited. An exclusion added before you know what the problem is exists to guard against a possibility rather than a result.

What Exclude will not do

It will not override the genre. If the style you named implies a sound, excluding that sound asks the model to serve two masters, and the genre usually wins.

It will not work the same way on every model. Songsmith reports that "v5/v5.5 handles negatives more reliably than older versions", which is plausible and unmeasured, so it is a reason to retest rather than a reason to trust.

And it will not tell you whether it did anything. There is no diff between a take with an exclusion and the take you would have got without it, which is why the honest way to use this field is one change per generation, so you can hear what moved.

Open Advanced Options, move your exclusions out of the Style box, and cut them down to the two that are actually about absence.

Read next

Two voices in one generation is contested, and the workflow that people report succeeding with builds each voice separately.

5 min read

Available now on iOS and Android

© 2026 Heho.ai. The professional lyrics workspace.