Transfer motion to a character with Wan Fun

Source comparison

These owned 9:16 sources are inputs to the observed Wan request. The five-second motion clip is a motion reference input, not the finished instructional video.

View the source rights receipt (for maintainers)

Wan Fun combines two kinds of guidance. A still image supplies the character and a short motion clip supplies the performance. In the reviewed capture, matching those inputs at 9:16 gave the model a clear starting point before one deliberate paid request.

Before you start: choose a clean character source

Use an image you own or have permission to use. The worked character reference is a fictional silver-haired elven ranger, 1080 by 1920 pixels. Keep the face, clothing, body outline and important details visible. Text overlays, a hidden face and a crop that cuts off active limbs give the model less reliable identity information.

The still answers who should appear. It guides the face, hair, clothing, colours and proportions. It does not provide the action. Keep a rights record for the image and do not assume that owning an illustration gives you permission to use somebody else's performance footage.

Match the frame

The owned motion reference is 720 by 1280 pixels and lasts five seconds. Its resolution differs from the still, but both sources are 9:16 portrait frames. That shared shape matters more than matching the pixel count exactly.

Aspect ratio is the shape of the frame. Resolution is the number of pixels within it. A 1080 by 1920 image and a 720 by 1280 video can work together because both simplify to 9:16. A landscape 1920 by 1080 clip asks the model to reconcile a very different crop.

Do not stretch either file to force a match. Crop safely instead, keeping the face, hands and active movement in frame. Similar subject scale and camera angle make the relationship clearer too.

Verify the live controls

The live capture was reviewed on 29 July 2026 with Wan 2.2 (Fun). The still was attached under Character reference, shown as Main after upload. The five-second clip was attached under Motion Reference. Confirm the model and these labels in your own account because availability and wording can change.

The observed price beside Generate was 20 Gold. That was a dated observation, not a price promise or an entitlement guarantee. Check the live cost and entitlement immediately before a paid action. Confirm both previews are the correct owned files before writing a prompt.

Run one deliberate test

Use a compact prompt that supports the references rather than contradicting them. For this capture, the intent was to keep the silver-haired ranger's face, clothing and proportions consistent while following the reference movement with steady full-body motion.

Press Generate once and wait for that original request to reach a terminal state. The reviewed capture recorded one completed request. Do not double-click, refresh into uncertainty or submit again because processing looks slow. If a paid request fails, record its visible status and confirm whether the original job is terminal before considering a new attempt.

Review cropping and motion fidelity

The one observed result was a 474 by 844 portrait video, just under five seconds long. It followed the portrait input shape rather than the disabled size display seen in the interface. Treat that as one reviewed outcome, not a guarantee for later requests.

Review identity first, then motion, framing, continuity and privacy. The face, hair, costume and body shape should still resemble the character reference. Major poses and timing should follow the motion input without frozen or duplicated limbs. Check that hands and active movement remain in the portrait frame, then watch for distracting changes in facial features, costume edges or the background.

If the result misses the target, change one cause at a time. Improve the character image when identity drifts. Trim the motion source to one readable action when the performance is unclear. Align aspect ratio, subject scale and camera angle before spending Gold on another variation.

Transcript

What Wan Fun transfers

Wan Fun transfers two different kinds of guidance into one generated video. A still image tells it who should appear. A short motion clip tells it what should happen. Used together, they give the model a character to preserve and a performance to follow.

Character reference and motion reference

The character reference is for identity. It guides the face, hair, clothing, colours and proportions. The motion reference is for action. It guides timing, pose changes, movement and camera behaviour. They are not interchangeable. A clear still with an obscured motion clip, or the other way round, leaves the model to guess at an important part of the result.

Resolution and aspect ratio

It also helps to separate resolution from aspect ratio.

Resolution is the number of pixels in a file. Aspect ratio is its shape, expressed as width to height. More pixels can preserve more detail, but they cannot fix a poor crop, a hidden face or an incompatible pose.

Our character image is 1080 by 1920. The motion video is 720 by 1280. Those are different resolutions, but both simplify to 9:16. They are both portrait frames. Matching that shape gives the two references a much better starting point than combining a portrait character with a wide landscape performance.

Reviewed live controls and paid boundary

For this reviewed capture, open Generate, choose Video, then select Wan 2.2 (Fun). Add the still under Character reference. After upload, that control appeared as Main. Add the short clip under Motion Reference. Check both previews before you write a longer prompt.

The live interface can change, so confirm the model and labels available to your account before following a paid path. The capture showed 20 Gold beside Generate. Treat that as an observed price on 29 July 2026, not a lasting price promise. Check the displayed amount immediately before submitting.

When you are ready, press Generate once and wait for that original job to reach a completed or failed state. Do not double-click, refresh into uncertainty, or submit again just because processing takes time.

Completed result and review

In this example, one submission completed. The result was a portrait video, 474 by 844 pixels and just under five seconds long. It followed the matching portrait sources rather than the disabled size display seen in the interface. That is why the completed output matters more than an assumed setting. Review what actually came back.

Start the review with identity. Does the face, hair, clothing, colours and body shape still resemble the character reference? Then review motion. Do the main poses and timing follow the motion clip without frozen or duplicated limbs? Check framing next. The subject and active movement should stay inside the portrait frame.

Then inspect continuity. Watch hands, facial features, costume edges and the background for distracting jumps between frames. Finally, review privacy. Make sure no unintended person, screen, address, message or private background from either source has appeared in the output.

Recover one change at a time

If something is wrong, change one cause at a time. For identity drift, improve the still or remove conflicting prompt detail. For unclear motion, trim the clip to one readable action. For framing problems, align the aspect ratio, subject scale and camera angle before trying a new prompt variation. Keeping each change separate makes the next result easier to diagnose and keeps a paid workflow deliberate.

Watch the vertical derivative, 56.68 seconds at 1080 by 1920.