Creating Videos with Character Consistency (v1.01)

When generating videos from a Start Frame, adding references helps the AI maintain consistent character appearance across subsequent sequences, shot transitions, and camera angle changes, especially when characters are viewed from behind, appear at a small scale, or are not clearly visible in the initial frame. This allows the AI to generate animations that stay faithful to the Start Frame while consistently following the provided references.
See also: Tutorial - Seedance Consistency with References

Start Frame

In this example, the video begins with one actor appearing at a very small scale in the Start Frame and gradually enlarging throughout the sequence. Compare the AI outputs with and without references to see the benefits of reference-driven character consistency, particularly when handling extreme scale changes, camera angle variations, or content generated from 3D previz scenes with gray-model characters or objects:

  • Stronger character consistency: Preserves facial features, skin details, facial silhouettes, and outfit colors and styles across shots and camera angles.
  • Higher-quality close-ups: Produces sharper and more detailed facial features, especially when the character is initially small, distant, or lacks sufficient detail in the source image.
  • Distinct character voices: Maintains a unique voice for each actor (in this example, lip-sync is driven by the assigned audio), improving character recognition and dialogue clarity.

AI Output without Reference AI Output with Reference

Since the actor appears at a small scale in the Start Frame, limited visual and audio detail can reduce the model’s ability to preserve character identity. This may lead to inconsistencies in character design and scale across generations, and less distinctive voices.

By relying on AI Actor references, character appearance and scale remain consistent, while audio-driven sources ensure a distinctive and recognizable voice.
1st Generation
2nd Generation
Prompt
  • Without an AI Actor reference, no actor tags are highlighted for Aaron and Tate, making them less clearly recognized by the AI and resulting in low character consistency in the generated video.
  • Without an audio reference, the characters’ voices lack a distinct identity.
  • With an AI Actor reference, actor tags such as Aaron and Tate are highlighted to help the AI recognize each character, ensuring high consistency in the generated video.
  • With an audio reference, adding "Lip-sync driven by [Audio]" to the prompt enables generation of distinct character voice audio.