Video
WAN 2.6
Video
Available on Ofrai — chat, images, and video in one place with shared credits.
Example prompts
Choose an example to load its prompt and recommended settings into the WAN 2.6 creator above.
How it works
Create with WAN 2.6 in three steps
- 01
Choose your inputs
Keep WAN 2.6 selected, then add a prompt and any supported reference assets.
- 02
Set the output
Choose the aspect ratio, resolution, and available Video options before generating.
- 03
Generate and refine
Review the result, adjust the prompt or settings, and generate another version from the same place.
Wan2.6 is one endpoint that never asks you to name a mode. The service reads the parameters you send and routes the request to the appropriate model tier, so the mode is a consequence of your payload rather than a field you set. Send a prompt and nothing else and you get text-to-video. Add an image through image_urls and you are in image-to-video. Put reference material in metadata.reference_urls and you are in reference video, also written r2v.
Three rendering paths, one endpoint, and the routing happens on the server. Everything else about this model follows from that shape: the reference fields and the aspect ratio are not available in every mode, the prompt is mandatory in exactly one of them, image input is limited to a single URL, and billing is per second with a different rate per resolution. The parameters you add are not only a creative decision.
One endpoint, three modes, and the parameters that choose
The routing has three documented outcomes and each has a trigger you control. No media and a prompt means text-to-video. Exactly one URL in image_urls means image-to-video. Reference material in metadata.reference_urls means reference video.
Those two fields are explicitly not combinable — the documentation states image_urls cannot be used together with metadata.reference_urls — so the routing is never ambiguous, but it is also not additive. You cannot send a source frame and a style reference in one call and expect both; you choose which kind of request this is.
Two other parameters confirm the same mechanism rather than documenting it separately: prompt is required in text-to-video and optional in the other two modes, and aspect_ratio is documented as applying to text-to-video and reference video, with image-to-video explicitly not supporting it. A model that routes by payload is a model where the shape of your request is the API.
The prompt is mandatory in exactly one of the three modes
In text-to-video, prompt is required. In image-to-video and reference video it is optional, and there it describes the expected action or the style rather than the content itself, because the content has already been supplied as media. The example given in the documentation is a plain content description — a cute cat stretching in the sunlight — with no camera grammar attached, which is a useful calibration for how much structure this field wants.
In image-to-video the prompt is effectively your only lever on motion, since the source image has already fixed subject and composition. That makes short deliberate instructions valuable: saying that the camera should hold still is real direction, not a wasted sentence. In reference video the same field is doing style work, which is a different job and worth writing for separately rather than reusing a text-to-video prompt unchanged.
image_urls takes one image, and only as a URL
The image field is an array, but image-to-video mode supports exactly one image. The format restriction is firmer than the array shape suggests: only publicly reachable http or https URLs are accepted, and base64 is no longer supported. Local images have to go through the image upload endpoint first to obtain a URL.
The field cannot be used together with metadata.reference_urls, which is the same mutual exclusion that decides routing — so in practice the single-image constraint and the reference-video path are two ends of one decision rather than two independent options. One consequence for pipeline design: because the request carries a URL and performs no upload, anything you need to render has to be hosted and reachable before the call. A job that dies on an unreachable local path is an integration bug, not a model behaviour.
aspect_ratio is not an image-to-video parameter
The aspect ratio defaults to 16:9 and takes 16:9, 9:16, 1:1, 4:3 or 3:4. It is documented as applying to text-to-video and reference video, and image-to-video is documented as not supporting it, because the source image governs the frame there. The trap is not that a value is rejected; it is that your own logs will disagree with the rendered output.
A pipeline that records 9:16, sends it, and assumes it was honoured will produce vertical deliveries that are actually whatever the input image was. Derive the effective ratio from the image in image-to-video mode instead of trusting the field, and keep the ratio where it does apply — text-to-video and reference video — where the five options give you genuine compositional control rather than a crop after the fact.
Non-Flash builds come with sound
Audio is an optional parameter that takes true, and the documented default behaviour is that non-Flash Wan2.6 output includes sound. That is the detail to plan around, because it inverts the usual assumption: you do not add audio to a Wan2.6 render, you remove it if you did not want it.
For a clip that will be cut against dialogue or a music bed, the default means an audio track arrives whether or not the prompt mentioned sound — the sound is generated, not requested, so its presence carries no information about the brief. The audio parameter exists to enable it explicitly, which is what you would do to make the intent visible in the request or on a build where the default does not apply.
Treat the default as a downstream obligation: either strip the track in post or plan the edit around what came back.
5, 10 or 15 seconds at 720p or 1080p, billed per second
Duration takes 5, 10 or 15 seconds and defaults to 5. Resolution takes 1080p or 720p and defaults to 1080p; 480p is not supported at all. Billing is per second and the rate differs by resolution, so the two parameters compound rather than merely configure: a 15-second 1080p render is three times the seconds of the default 5-second 1080p render, and a 15-second 720p render is a different line item again even at the same length. That is what shapes the working loop here.
Draft at 720p and 5 seconds, where a mistake is cheap, and spend the resolution and the length only once the shot is settled. The coarse duration ladder also means shots should be planned in five-second blocks; asking for eight is asking for a model parameter that does not exist.
FAQ
- How does Wan2.6 decide which mode to run?
- The server routes from the parameters you send rather than from a mode field. A prompt with no media is text-to-video, a single URL in image_urls is image-to-video, and reference material in metadata.reference_urls is reference video, or r2v. image_urls and metadata.reference_urls cannot be used together, so the choice is exclusive.
- Can I send more than one image?
- No. Image-to-video supports exactly one image. It must be a publicly reachable http or https URL — base64 is no longer supported, so local files go through the image upload endpoint first to get a URL — and the field cannot be combined with metadata.reference_urls.
- Does Wan2.6 produce audio?
- Yes, and by default on non-Flash builds: the documentation states that non-Flash Wan2.6 outputs video with sound unless told otherwise. The optional audio parameter takes true to enable it explicitly, which is useful for making the intent visible rather than for switching a silent build on.
- What resolutions and durations exist, and how is it billed?
- 720p and 1080p only, with no 480p, and durations of 5, 10 or 15 seconds defaulting to 5; 1080p is the default resolution. Billing is per second and the rate differs per resolution, so length and quality multiply — draft at 720p and 5 seconds, then spend both at the end.
FAQ
- Can I use WAN 2.6 directly on this page?
- Yes. The creator above is the working Ofrai studio with WAN 2.6 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 2.6?
- 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.
Related models
Compare other Video models that can handle similar creative work to WAN 2.6.
Compare
Use cases
Related guides
Tools
Commercial use depends on the selected model and model terms.





