Wan 2.2 14B Move: the still is the contract (720p, 8cr)

Wan 2.2 14B Move is image-to-video. If the label is wrong, every second is waste.

The catalog category is more specific than the I2V label: motion-control, 14B parameters, 8 credits, 720p, qualities 480p, 580p, and 720p. It requires an image. Catalog audio is unset — do not budget a talking generate. There is no duration enum on the card. The motion does not come from a paragraph. It comes from a driving clip plus the still that is the identity.

Call this image-to-video only if you understand the extra file. If you only have a prompt, you are on text-to-video. If you only have a still, you are on ordinary I2V. If you have a still and a performance, you are on Wan 2.2 14B Move.

The still names who; the clip names when

Two required uploads:

1. Character / reference image — the identity. Framing matched to the performance. This is the contract the title is about. A headshot against a full-body drive is how the transfer looks like a different person.
2. Driving video — the motion source. Walk, turn, gesture, dance. Rights to use it. Subject in frame.

The prompt is optional guiding text. Wan's family tip is straightforward scene descriptions and style keywords. "Studio grey, red jacket" is in register. "Then she pirouettes, cinematic crane, epic" is arguing with a file. The motion-control job copies timing you already captured. Text cannot replace the clip. That argument is also motion control needs a driving clip.

Mislabel the row as "just I2V" and you will either fail the schema (no video_url) or silently invent motion that is not the performance you paid to keep. Every second of that output is waste relative to the brief.

720p is the top of this card

Max output resolution is 720p. 480p and 580p exist as lower steps. The 8-credit figure is the HD/4K-labelled option on a per-second matrix; SD is listed at 4. There is no native-audio promise. If the driving clip has a score, do not assume the transfer keeps it. Finish sound after, or pick a talking/lipsync model when the mouth is the product.

AI motion transfer is the tool-shaped door. Wan's roster is Alibaba's video family; this 2.2 14B Move card is the motion-control member, not Wan 2.7 text-to-video. The best motion-control model ranking is the rest of that shelf.

When the still is wrong, stop

Do not "fix identity in the prompt." The still is the contract. If the character is off, regenerate the still on a text-to-image row, match the crop to the driving clip, then run Move. Spending 8 credits to transfer a bad likeness is how you buy a long, expensive almost.

No duration list means you do not get to type "8 seconds" into the prompt and expect a trim. The driving clip's length is the performance. Trim that file first.

FAQ

Is this ordinary image-to-video from one still?

No. It requires the still and a driving video. One still plus a motion paragraph is a different family. If the label on the job is "animate this photo," do not open 14B Move.

Why is the still the contract if the clip does the moving?

Because identity is sampled from the image. The clip only donates timing. A wrong still makes every transferred frame the wrong person, at 720p, at 8 credits.

Does Wan 2.2 14B Move generate a soundtrack?

The catalog does not list audio on this row. Do not write dialogue into the optional prompt and expect a voice file. Plan sound after the transfer, or use a lipsync row when the mouth is the shot.

What resolution should I pick?

The card tops out at 720p, with 480p and 580p as lower qualities. Pick 720p when the delivery is the transfer. Pick lower when you are checking the framing match before a keeper pass.