Wan 3.0: Give Every Reference File One Job

You have three useful files—a stall photo, a clip of bunting moving in the wind, and a market recording. Use all three in the same video without a plan, and the photo may reshape the setting, the motion may spread to the wrong objects, and the recording may introduce voices you never meant to use.

When using Reference to Video in Wan 3.0, decide what each file should contribute before uploading it.

Make a Three-Line Reference Map

For a short makers’ market teaser, the plan could be:

• Stall photo: table layout, canopy colour, and warm visual palette

• Bunting clip: gentle fabric movement only

• Market recording: quiet atmosphere without clear speech

A reference file shows Wan 3.0 what material to use. The prompt explains which part matters.

These limits matter because a reference can influence more of the scene than you intended. Leave any file without a clear purpose out of the first draft.

Give the Clip One Job

The video does not need to explain the whole market. It only needs to remind people that the event is coming.

For this market post, a simple 15-second sequence is enough:

1. Open with a wide view of the stall.

2. Let the bunting move gently above the table.

3. Move closer to the handmade items.

4. End on a steady frame with space for event details.

Dates, prices, directions, and registration information can be added later. Trying to fit them into the generated scene makes the clip harder to control.

Write the Prompt Around the References

Keep the prompt close to the reference map:

Make a 15-second vertical teaser for a weekend makers’ market. Keep the stall layout and warm colours from the photo. Use the video only to guide the gentle bunting movement. Make one slow camera push and leave space on the right for event text. Do not add new signs, products, prices, or crowds.

Focus on the decisions the references cannot make on their own: what to keep, what to limit, and what not to add.

Build the Draft in Layers

In Wan 3.0, choose Reference to Video and begin with the stall photo and bunting clip. Select a vertical format and a 15-second duration, then generate the first visual draft without the market recording.

Check the basic direction before adding anything else:

• Does the stall still resemble the photo?

• Is the bunting movement gentle?

• Does the camera make one clear move?

• Is there enough empty space for text?

If the visuals work but the scene feels empty, add the market recording to the next version. Ask for low background ambience without recognisable conversations.

If the table changes, revise the photo instruction. If the movement is too strong, adjust the bunting reference or motion wording. If the sound becomes distracting, reduce or remove it.

Change the instruction connected to the problem instead of rewriting the full prompt.

Review Four Specific Details

Watch the finished draft more than once, with one question in mind each time:

• Stall: Have the table, canopy, or products been replaced?

• Bunting: Is only the fabric moving, or are nearby objects moving too?

• Audio: Can you hear a clear conversation, song, or announcement?

• Facts: Has the video added a sign, price, vendor, product, or crowd that will not be at the real event?

These checks matter because the clip represents a real place and event. A small invented detail can create the wrong expectation.

Add Confirmed Details During Editing

Add the event name, date, address, price, logo, and call to action after the footage is ready. This keeps spelling and layout consistent and prevents generated text from changing between frames.

Treat the footage as a creative preview, not proof of what the event will look like. Use only photos, videos, and recordings you have permission to reuse. Take extra care when faces or voices are identifiable, and do not upload private client files that are unnecessary for the task.

Start with one image and one motion reference. Add another file only when you can name the problem it needs to solve.