Wan 3.0 Video Generator for Long Takes and Reference Images

Length is set in whole seconds anywhere from 2 to 30. Half a minute gives a scene time to breathe: a chef plating a dish from first ingredient to garnish, a camera drifting through a shop, a character walking into frame and reacting, all inside one generation instead of several stitched together.

Upload as many as ten pictures of the people, objects or places that belong in the scene. The model builds the clip with them in hand, which is the practical way to keep a mascot, a face or a bottle label recognizable across a long take, rather than relying on a written description alone.

Every uploaded picture is named by its position: the first one is Image1, the second Image2, and so on. Writing 'the dancer from Image1 wears the jacket from Image3' tells the model exactly which picture you mean, which starts to matter once there are more than two or three on the board.

Prefer to fix the composition yourself? Upload a still and the clip opens on it. If you already know where the shot should land, add a second still as the closing frame, for instance a gift box shut at the start and open at the end, and describe what happens in between.

Audio is generated in the same pass as the picture, and the toggle starts in the on position. Keep it on when you want the clip to arrive with sound. Switch it off when an editor will lay music or a voice-over on top later, so there is nothing to strip out first.

The adaptive setting takes its shape from what you feed in, so an upright photo stays upright. When the destination decides the frame instead of the source, choose 16:9, 9:16, 1:1, 4:3 or 3:4 directly.

Upload the front, the back and the logo of a product as references, then describe how it is unboxed, poured or worn. Naming each angle in the prompt helps the label and shape stay close to the real item for the whole clip.

Keep one set of character pictures and reuse it in every episode of a short series. The same Image1 in each prompt is a steadier starting point than describing the character's face and clothes from scratch each time.

Use one storyboard panel as the first frame and the next as the last frame, then describe the move between them. It is a quick way to see whether a transition works before anyone books a shoot.

Set 9:16 and a length close to 30 seconds for short-form feeds, when one unbroken shot can carry a recipe, a trick or a before-and-after without fast cutting.

A slow camera path through a café, a hotel room or a showroom needs time. Feed in photos of the space as references and describe the route the camera takes from the entrance to the window.

Agencies and freelancers can block out a mood film at 480P to agree on tone with a client, then render the approved version at 1080P with the same prompt and the same references.

Before uploading, decide what each picture is for: Image1 the presenter, Image2 the sneaker, Image3 the street corner. Then say so in the prompt. Two near-identical photos of the same thing are hard to tell apart in writing; one clear photo per element is easy to name.

Names follow upload order, so put the main character first if they appear first in your text. If the order comes out wrong, re-uploading in the right sequence is usually quicker than rewriting the prompt around the numbers.

For 20 or 30 seconds, list what happens in order: the opening view, the main action, the turn, where the camera ends up. A one-line prompt stretched over half a minute leaves most of that time unaccounted for.

Starting from a first frame or from references with a strong composition? Adaptive keeps that shape. With text only, or when a platform needs a fixed format such as 9:16 for vertical feeds, set the ratio yourself.

The price climbs with both length and resolution, so test the idea at 480P and a short duration first. Once the motion and timing look right, run the same prompt again at 1080P and the full length you need.

Say what the subject does in one sentence and what the camera does in the next, for example 'She pours the coffee. The camera slowly pushes in from the doorway.' When only one of the two needs changing, you know exactly which line to edit.