Decide which part of the scene to control
Text mode lets you describe a scene idea. First and last frames can define the shot’s boundaries, while reference mode lets you work with existing material. These modes expect different inputs, so choose the workflow before uploading files.
Wan 3 generation settings
- Output duration
- 2–30 seconds, selected in the generation form.
- Resolution
- 480P, 720P or 1080P.
- Input modes
- Text, first/last frames and reference mode.
- Aspect ratios
- 16:9, 4:3, 1:1, 3:4, 9:16 or adaptive.
- Audio and watermark
- The current workflow requests output with audio and without a watermark.
- Reference video budget
- Combined reference video duration and output duration must not exceed 30 seconds.
Reference video shares the output time budget
In Wan 3 reference mode, duration is not just an output setting. The combined duration of reference videos plus the generated output must not exceed 30 seconds. Reference videos also have their own separate 15-second combined limit.
For example, 12 seconds of reference video leaves room for an output of up to 18 seconds. Choosing a 30-second output leaves no time for reference video within that shared budget. Images and other supported inputs are different choices; removing a reference video does not start a generation by itself.
Keep frame inputs separate from the reference list
Frame mode uses the images that start and end your shot. In reference mode, you can add video or audio alongside an image; document and web-link fields are also available. Selecting the mode is not enough: add an actual reference.
Explain the source’s role in your prompt. If an image defines a character’s appearance, say so. If a video guides camera motion, describe the movement you want to retain. The form waits for file preparation and duration validation before submitting.
Plan portrait or landscape framing at the start
Portrait framing for a phone screen and landscape framing for a wide scene lead to different compositions. Choose the aspect ratio and resolution in the Wan 3 form, then set the duration. If using reference video, check the remaining time budget again.
Review the current credit cost before submitting. Completion time varies with workload. Watch the result in the generation screen when it is ready, then download it.
Questions about Wan 3
- Can I make a video transition between two images?
- Use first and last frame mode to set the images as the beginning and end of the shot. Describe what happens between them, such as a character turning to look out of a window. Very different lighting or composition can make the transition look unnatural.
- How do I move just the water or clouds while keeping the rest still?
- Limit the movement to a specific element: “Locked camera; only small ripples move across the water.” Avoid adding a pan or zoom to the prompt. The model may still change other parts of the image.
- Will adding a reference video give me an exact copy?
- No. A reference guides a new generation rather than copying the original. Explain what you want from it: camera motion, a character’s action or the atmosphere. Check the finished clip for details you wanted to preserve.
- Why can’t I choose a longer output after adding a reference video?
- Reference videos and the new output share a 30-second budget. A 12-second reference, for example, leaves at most 18 seconds for the output. If you need a longer result, use only the relevant part of the reference.
- What can I try if the image changes abruptly between the first and last frames?
- Try two images with more similar camera angles and lighting. Split a major location or clothing change into separate shots instead of forcing it into one transition. Start by describing one main action between the frames.