Wan 3.0 Uncensored Video API: All Six Endpoints Are Live

Key Takeaways

• All six Wan 3.0 endpoints are live on Siray: alibaba/wan-3.0-t2v, -i2v, -ref2v, and the Spicy variant of each.
• The Spicy and standard endpoints share one request body. Excluding the model string, their published schemas are field-for-field identical.
• Every Wan 3.0 endpoint accepts durations from 2 to 30 seconds and renders at 480p, 720p, or 1080p.
• Reference-to-video takes up to 10 images, 5 videos, and 5 audio files. No Wan 3.0 endpoint accepts a document or text file as input.

What's Live Now

Wan 3.0 ships on Siray as six endpoints across two tiers. Wan 3.0 itself covers text-to-video, image-to-video, and reference-to-video; the Spicy tier mirrors all three for mature, boundary-pushing work.

The Spicy tier's arrival was covered in detail here. What has changed is that the full lineup is now callable rather than partially documented.

Spicy and Standard Share One Request Body

This is the part that matters for integration work. Comparing the published specifications for each matched pair — t2v against t2v-spicy, and so on through i2v and ref2v — the request bodies are identical once the model string is set aside. Same path, same required fields, same enums, same optional parameters.

The consequence: moving a working standard integration to the Spicy tier is a one-string change. There is no separate adult API and no extra parameter to discover. The same pattern holds across the wider pool, as the job submission model already demonstrated.

What Every Request Must Include

Five fields are required on all six endpoints: model, prompt, duration, size, and aspect_ratio. Two endpoints carry additions:

• i2v and i2v-spicy additionally require image, and optionally accept end_image to pin the closing frame.
• ref2v and ref2v-spicy accept images, videos, and audios arrays instead of a single source frame.

aspect_ratio is required even on image-to-video, where it might look redundant — the source image does not imply it. Accepted values are 16:9, 9:16, 1:1, 4:3, 3:4, and adaptive.

Duration and Resolution

duration is an integer count of seconds and accepts every value from 2 through 30. size accepts 480p, 720p, and 1080p. Both are required, so neither has a silent default to fall back on.

Thirty seconds in a single call removes most of the stitching work that shorter-ceiling models force. For pacing a prompt across a long take, the Spicy prompt guide breaks down how to distribute description across the full duration.

Reference Inputs, and What Is Not Accepted

The reference-to-video endpoints take up to 10 images, 5 videos, and 5 audio files per request, each supplied as a URL or a data URL. All six endpoints also expose audio_enable, a boolean that turns on audio generation alongside the video.

Worth stating plainly, because the opposite claim circulates: no Wan 3.0 endpoint accepts a document, PDF, or text file as input. The published inputs are prompts, images, videos, and audio. Nothing else.

Developer Notes

Wan 3.0 jobs are asynchronous, and the shape of the URL is not a reliable signal of that. seed accepts integers from -1 upward for reproducible runs. For running many jobs at once, see scaling batch generation.

FAQ

Do I need a different API key for the Spicy endpoints?

No. The same key and the same endpoint path serve both tiers. Only the model string changes.

What is the longest clip Wan 3.0 can generate in one call?

Thirty seconds. duration accepts any integer from 2 to 30.

Can Wan 3.0 generate audio?

Yes. Every endpoint exposes an audio_enable flag, and the reference endpoints additionally accept up to five audio inputs.

Summary

All six Wan 3.0 endpoints are live, and the Spicy tier sits one string away from any working standard integration. Create your free Siray account and start building with alibaba/wan-3.0-t2v-spicy today.

Generate with Wan 3.0 Spicy now

Siray refuses all illegal content and applies zero tolerance to CSAM. Every workflow described here is intended for legal, adult (18+) content only. "Spicy" describes working within legal moderation bounds, not a promise of unfiltered output.