Asset Selection Guide

Video generation supports reference images, videos, audio, and other assets. This page uses comparison cases to explain how to prepare and upload assets for better generation results.

When uploading assets, merging the target face image, the full-body reference image, and detail reference images into a single picture can shrink each reference element's share of the frame, making recognition harder for the model — causing the generated person to deviate from the uploaded assets, or triggering risk-control interception.

Upload the face close-up, costume details, and other key content as separate images.

Case Comparison: 3D Animated Parent & Child

Input: background reference image + costume three-view sheet + neutral-expression face close-up + prompt

Prompt:

3D animation style, background follows reference image 1. Character A (costume follows reference image 2; facial features strictly follow image 3) and Character B (costume follows reference image 4; facial features strictly follow image 5) walk hand in hand along a garden path, camera behind them. The camera cuts to the front: Character A picks a flower and gently hands it to Character B. Background SFX: soft wind stirring leaves and blossoms, birdsong. Character B smiles, takes the flower, smells it, then crouches down to meet Character A's eyes with a smile, gently pats Character A's head, and says: "thank you". Background SFX: a soft chime.

Case B

Input: background reference image + costume three-view sheet + prompt (face close-up missing)

Case C

Input: background reference image + costume front view (not a three-view sheet) + prompt

Comparison

  • Case A (three-view + face close-up): best facial fidelity.
  • Case B (three-view, no face close-up): weaker facial consistency.
  • Case C (front view, not three-view): weaker consistency in both costume and face.

Summary

Asset combinationResult
Background + costume three-view + face close-upBest
Background + costume three-view (no face close-up)Poor facial consistency
Background + costume front viewPoor in both costume and face

To use the platform asset library, contact the SilvaMux team for access instructions.