Skip to content

Video

WAN 3.0 Prime

Video

Available on Ofrai — chat, images, and video in one place with shared credits.

206 creditsModel details

Video studio

Preview

No clip yet

Your latest clip preview will show here after generation.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the WAN 3.0 Prime creator above.

How it works

Create with WAN 3.0 Prime in three steps

  1. 01

    Choose your inputs

    Keep WAN 3.0 Prime selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Video options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

Wan 3.0 Prime is documented as an all-purpose video generation surface: text-to-video, control from a first frame or from a first and last frame, combined generation from reference images, reference videos and reference audio, and reference to file or public web page content. The parameter that separates it from the shorter-window models is duration.

Output length runs from 2 to 30 seconds, and the option list includes -1, which hands the length decision to the model — it reads the prompt and the supplied media and decides how long the clip should be. That is a genuinely different tool from specifying a number, and it is worth deciding deliberately when to use it. Resolution is a three-rung ladder from 480p to 1080p with 1080p as the default, and audio generation is on unless you say otherwise.

A two-to-thirty-second window, and a -1 that gives the decision away

Duration takes any value from 2 to 30 seconds, and the documented default is 5. The -1 is the entry that deserves the attention: it is a value you pass, not the absence of a value, meaning the model chooses the output length itself from the prompt and the assets. That distinction matters more than it looks, because a request that omits duration gets five seconds and a request that passes -1 gets whatever the model judges the material to need — two completely different behaviours behind one field.

Thirty seconds is long enough for a single generation to cover a complete short scene rather than a single shot, which is exactly where -1 is most useful and most dangerous. In the right hands it saves you from guessing a length that fits the material; in a film edit it hands the edit to the model. If a downstream system has cut points planned, name the number.

prompt is optional, but not optional on its own

The prompt field is not required. The documented rule is that the prompt and the media assets need at least one between them, which produces a validation shape that is broader than most pipelines assume. A request carrying only an image is valid. A request carrying only reference video is valid. A request carrying only a prompt is valid. A request carrying nothing is not.

That single condition is the whole precondition, and it is easy to get wrong in both directions: enforcing prompt unconditionally rejects perfectly good asset-led jobs, while checking only that at least one field is present will happily accept a prompt full of placeholder text. Validate the union, not the intersection, and then let the asset-carrying requests skip prompt-specific prompt engineering entirely.

Referring to your own assets by array position

When references are used together, the prompt can address them by position: write image 1, video 1 or audio 1, following the order of the arrays. This is slot addressing, and it means the prompt refers to your inputs rather than describing them a second time — the same trade as any reference-driven model, where re-describing an asset in prose creates a competing description rather than a clearer one. The practical consequence is that array order is load-bearing.

Re-sort a reference list and every index in the prompt now points at different media, with nothing in the request to indicate that anything changed. Treat the ordering of the reference arrays as part of the prompt, not as an implementation detail of the payload: when a render comes back with the wrong subject, check whether the indices still resolve to the media you meant.

video_list: five clips inside a fifteen-second budget

Reference video goes in video_list, which accepts at most five entries with a combined duration of fifteen seconds or less. Each entry is a small object carrying a video_url pointing at the clip. The two ceilings are different in kind and worth holding separately: five is a hard count limit on the array, fifteen seconds is an aggregate across all of it, so five five-second clips exactly fills the budget and a single twenty-second clip does not fit at all.

This budget is separate from the output duration, and the two are easy to conflate — 15 seconds of reference material and a 30-second output are independent limits, so asking for smart duration with a full reference list is asking the model to stretch fifteen seconds of source into a longer result on its own judgement.

480p is reachable here

Resolution takes 480p, 720p or 1080p, with 1080p as the documented default. Having a genuine low rung is what makes this endpoint pleasant to work in: 480p is a real draft tier for iterating on timing and framing, 720p is a middle, and 1080p is the deliverable.

The default being the top of the ladder rather than the bottom is worth overriding deliberately, because a draft that inherits the documented default is a 1080p draft, and that is exactly the cost you were trying to avoid while still searching for the right shot. Set 480p explicitly while the prompt and the framing are moving, then move up once, rather than paying the top rung for every version you throw away.

Audio on by default, and a capability described in one line

The audio parameter controls whether audio is generated, and it is enabled by default. Same operational consequence as elsewhere in this family: a clip generated for a cut that will carry its own dialogue arrives with generated sound attached, so the audio switch is the first thing to set deliberately rather than discover in post. The capability list also includes reference to file or public web page content.

That is documented as available, but it is described in a single line with no parameter of its own in the same breath as the others, so verify the exact payload shape against the current reference before building on it rather than assuming it mirrors video_list. The documented detail is short; the safe engineering move is to treat it as unverified until you have seen it work.

FAQ

What does duration -1 do?
It is a smart-duration mode: the model chooses the output length from the prompt and the supplied media. It is a value you pass, not a missing field — the documented default when duration is omitted is 5 seconds, so -1 has to be sent explicitly to get the model's judgement instead of the default.
Is the prompt required?
No. The rule is that the prompt and the media assets require at least one between them, so a request carrying only an image or only reference video is valid, as is one carrying only a prompt. Validate the union of those fields rather than requiring a prompt unconditionally.
How many reference videos are allowed and for how long in total?
video_list accepts at most five entries with a combined duration of 15 seconds or less, each entry an object with a video_url field. The five-entry count and the fifteen-second aggregate are separate limits, and this reference budget is independent of the 2 to 30 second output duration.
Which resolutions are available, and what is the audio default?
480p, 720p and 1080p, with 1080p as the documented default — so set 480p explicitly while iterating rather than letting drafts render at the top of the ladder. Audio generation is enabled by default and is controlled by the audio parameter.

FAQ

Can I use WAN 3.0 Prime directly on this page?
Yes. The creator above is the working Ofrai studio with WAN 3.0 Prime selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with WAN 3.0 Prime?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Video models that can handle similar creative work to WAN 3.0 Prime.

Commercial use depends on the selected model and model terms.