Build a Reference Map for One Wan 3.0 Scene

It is a short record of which uploaded source controls each production decision. An image can own identity, a video can own a movement pattern, and an audio file can own timing. The prompt connects every role with the matching @ token so the result can be reviewed against the same map.

Image to Video uses one opening frame or a first-and-last-frame pair as the main visual constraint. Reference to Video is useful when separate images, clips, and audio files must own different decisions inside one scene.

Standard Wan 3.0 multimodal reference mode accepts up to 10 images, 5 video clips, and 5 audio files. Those are limits, not targets. A smaller set with distinct roles is usually easier to direct and diagnose.

Upload the source, type @ in the scene brief, select its token, and place it beside the responsibility it owns. For example: preserve the subject from @image1, follow the gesture from @video1, and land the visible action on the cue in @audio1.

Choose first and last frames when the opening and destination are the two decisions that must stay fixed. Choose multimodal references when identity, motion, camera language, or sound come from separate sources.

Standard Wan 3.0 is preselected because it supports image, video, and audio references together. The model picker remains available; confirm the active reference types, duration, resolution, audio controls, and credit estimate before submitting.

List the source filename or token, its one primary responsibility, the visible evidence you expect, and its priority when another source overlaps. Then write the scene in chronological order. If a source has no unique responsibility or no observable success condition, leave it out of the first pass.

Choose one image as the authority for every important visual category: subject identity, product shape, wardrobe, environment, palette, or composition. State that authority beside its @ token, remove near-duplicates, and decide which source wins when colors, proportions, or character details disagree. The limit is 10 images, but a smaller map is easier to audit.

Yes. Attach the clip with its @ token and name only the property to borrow: gesture, blocking, camera speed, transition timing, or action rhythm. Define the new subject and environment with image references or text. Standard Wan 3.0 accepts up to 5 reference videos; if clips demonstrate different versions of the same move, name the priority.

Give the audio a named function—dialogue cadence, musical beat, room ambience, or effect—and tie a key sound to a visible event on the timeline. Standard Wan 3.0 accepts up to 5 audio files. Review once with sound to check timing, then without sound to confirm the visual action still communicates.

Use one line per assigned role: identity from its image, movement from its clip, and timing from its audio. Mark each as clear, partial, or missing, then inspect continuity, camera behavior, background stability, fine detail, ending state, and synchronization. Revise the weakest role before changing unrelated settings.

Record the model, prompt, duration, resolution, audio setting, source filenames, @ mapping, and assigned roles. Name each result by the one controlled change, such as “identity-image-B” or “slower-motion-reference.” Keep the same scorecard across versions so reviewers compare the tested source rather than unrelated differences.

Identify the exact category in conflict—identity, product shape, color, movement, camera, or timing—then choose one source as the authority for that category. Remove any source with no separate job. If motion fights the still composition, test a smaller move before adding more instructions or assets.