Start
Start
Documentation Menu

Character Consistency Workflow

Generate multiple video scenes featuring the same character with consistent appearance using reference images and the @tag system.

Difficulty

Intermediate · ~15 minutes · Nodes: Text, Reference Image × 3, Scene × 2+

Ingredients

  • 1 Text Node — Character description
  • 3 Reference Image Nodes — Character views (front, side, back)
  • 2+ Scene Nodes — Video scenes with the character
  • Global Context enabled on all generation nodes

Procedure

Step 1: Create the Character Description

Add a Text Node in Static mode with a detailed character description:

“A young woman named Luna with long silver hair, emerald green eyes, wearing a blue adventurer’s cloak with gold trim. She carries a wooden staff with a glowing crystal. Anime art style, detailed, consistent design.”

Set this Text Node’s tag to @character-desc.

Step 2: Generate Reference Images

Add 3 Reference Image Nodes. Connect the Text Node to each one’s input port. In each Reference Image’s prompt, add view-specific instructions:

  • Ref 1 (tag: @luna-front): “Full body front view, standing pose, @character-desc”
  • Ref 2 (tag: @luna-side): “Full body side profile view, walking pose, @character-desc”
  • Ref 3 (tag: @luna-back): “Full body back view, looking into the distance, @character-desc”

Generate all three. Pick the best result for each, and lock them to prevent re-generation.

Step 3: Create Video Scenes

Add Scene Nodes for each video clip you want. Connect ALL 3 reference images to each Scene as visual references (up to 3 per scene). In each Scene’s prompt, reference the character tags:

  • Scene 1: “Luna (@luna-front) walks through a magical forest, soft sunlight filtering through trees, gentle camera movement”
  • Scene 2: “Luna (@luna-side) discovers a glowing lake, she kneels down to look at her reflection, cinematic”

Step 4: Enable Global Context

On each Scene Node, make sure “Use Global Context” is enabled. Add the character description Text Node as a global context source. This ensures every scene “knows” about Luna’s appearance.

Step 5: Generate and Iterate

Run the workflow. Review each scene for character consistency. If a scene doesn’t match:

  1. Lock the good scenes (click the lock icon)
  2. Adjust the prompt for the inconsistent scene
  3. Re-run — only unlocked scenes regenerate

Result

A series of video clips featuring the same character with consistent appearance across different scenes and settings.

Variations

  • Add a Draw Node before the Reference Images for precise pose guidance
  • Use If/Else to branch into different settings (indoor vs outdoor)
  • Connect AI Voice Node for character narration
  • Use Upscaler on the reference images for higher quality source material
  • Add more Reference Images (different expressions, outfits) for variety while maintaining consistency

See Also