Wan 2.7 Spicy Audio: Adding Sound to NSFW Video Clips

Silent adult clips are half a deliverable. Most coverage of NSFW video generation stops at resolution, duration and prompt wording, skipping what comes out of the speakers. On Siray, wan-2.7-i2v-spicy exposes two separate audio fields, and they do different things.

Key Takeaways

• audio takes an input audio URL — a track supplied by the user
• audio_enable is a boolean that asks the model to generate audio alongside the video
• These are two distinct paths, not two names for one switch
• duration is every integer from 2 to 15 seconds, which sets the outer bound on any track
• The schema documents an audio input; it does not document lip-sync

Two Fields, Two Jobs

The request body for wan-2.7-i2v-spicy carries both fields, described in the published schema as follows:

That distinction is the whole article. One field accepts something already recorded and hosted somewhere reachable; the other asks the model to produce sound itself. A workflow that assumes audio_enable will apply a supplied file has the mental model backwards.

When to Supply Your Own Track

Bringing a track means control over exactly what plays: recorded voice, licensed music, room tone, or a mix bounced from an editor. For creators working to a consistent style across a catalogue, it keeps every clip sounding like one production.

It also puts the licensing question where it belongs: supplied audio has known rights. Generated audio does not.

The practical requirement is hosting: audio takes a URL, so the file has to sit somewhere the API can fetch it.

When to Let the Model Generate

audio_enable is the low-effort path. It fits clips where the sound is ambience rather than content — the background of a scene, not a performance. It is also the only option when no recording exists and none is going to.

The Duration Ceiling Cuts the Track, Not the Other Way Round

duration is a required integer with an enumerated range of 2 through 15, every value in between available. size is required too, and offers 720p or 1080p.

This matters more for audio planning than it first appears. A supplied track has to be trimmed to the clip length before it is uploaded, because the clip length is fixed at request time and cannot stretch to accommodate a track that runs long. Fifteen seconds is the hard ceiling on this endpoint. Anything longer is a stitching job in an editor, not a generation parameter.

Working backwards from the intended cut is the cleaner habit: decide the clip length, trim the audio to exactly that, then set duration to match.

What the Schema Does Not Promise

This is worth being blunt about, because the gap between "there is an audio field" and "the audio is synchronised" is where expectations break.

The published schema describes audio as an input audio URL. There is no field for lip-sync, no field for phoneme alignment, and no description text anywhere in the request body that mentions mouth movement or synchronisation. Treating the audio input as a lip-sync feature is an assumption, not a documented behaviour.

The same caution applies to quality. The checks behind this article read the request schema; they did not evaluate generated output. Field availability is confirmed; how the audio sounds is a question a schema cannot answer.

Where the Rest of the Controls Live

The audio fields sit alongside the ones that do the heavy lifting elsewhere: negative_prompt, a seed running from -1 to 2147483647, and prompt_expansion_enable, which defaults to true.

For the endpoint's general capabilities and its place in the family, Wan 2.7 Spicy i2v covers the broader picture, and the Wan 2.7 uncensored prompt guide handles prompt construction. For a model built around audio from the ground up rather than exposing it as two fields, MiniMax H3's native audio is the adult-adjacent alternative — a different moderation posture, and a different tool.

Create your free Siray account and start generating uncensored video with sound on Wan 2.7 Spicy.

A Note on Compliance

Siray's NSFW models generate legal, consensual adult content only — CSAM and any illegal content are refused with zero tolerance. Uncensored means no excess moderation within the law, not the absence of law. Every example here assumes adult subjects, and any supplied image or audio track is assumed to be used with the consent of the people in it and with the rights required to use it. Wan models are third-party models available through Siray's API, not Siray-hosted models.