GEMINI OMNI PROMPTS

CHARACTER

16:9 10s Reusable character workflow

character

Reusable Character: @character_name Multi-Shot Continuity

Use Gemini Omni's reusable character ID system to maintain identity across multiple shots — verified by Medium's Gemini Omni Prompt Playbook.

character reusable-id multi-shot omni-exclusive consistency

Prompt

Setup — Define the character first (one-time, with reference image upload):
Create a character named @maya. She is a 30-year-old woman with shoulder-length
curly brown hair, warm hazel eyes, and a soft natural smile. She wears a beige
wool sweater and dark jeans. Save this as a reusable character.

Generate — Use @maya in a multi-shot sequence:
@maya enters her morning routine. Shot 1: pours coffee from a French press at her
kitchen counter. Shot 2: opens her laptop at the dining table and starts typing.
Shot 3: walks to the window holding her coffee, looks out at morning fog. Hold
consistent face, hair, outfit, and lighting across all three shots.

Why this prompt

The @character_name reusable identity system is a Gemini Omni-exclusive feature — Sora and Veo cannot maintain character identity across separate clips this way.

Pattern adapted from Medium’s Gemini Omni Prompt Playbook, which validates this exact multi-shot continuity workflow.

Source tier: 🟡 Community-verified playbook (medium confidence — pattern documented, output quality varies)

How @character_name actually works

Per Google’s announcement and DeepMind’s docs:

  1. First turn: define the character with a description + (optionally) a reference image
  2. Save as reusable ID: Omni stores the character signature in your session
  3. Subsequent prompts: reference with @character_name instead of re-describing

This lets you generate dozens of clips featuring the same person across different scenes, poses, and shots — without identity drift on each generation.

Known limitations (be honest about this)

Per Atlas Cloud’s multi-turn consistency review, character consistency scored 3/5 in their testing. Specifically:

  • ✅ Hair color, eye color, general face structure: stable
  • ⚠️ Fine facial details, hand articulation, clothing textures: drift across shots
  • ⚠️ Across 4+ multi-turn iterations, drift compounds noticeably

Bottom line: reusable characters work well for 2-3 shots in one session, less reliably beyond that.

How to tweak

  • Add a reference image when defining the character (much more stable than text-only)
  • Be specific about identifying features: “curly brown hair to shoulders, beige sweater, small silver hoop earrings” instead of “brown hair, casual”
  • Hold-list explicitly: “Hold consistent face, hair, outfit, and lighting”
  • Reset on drift: if shot 4 looks wrong, regenerate from shot 1 with refined prompt

Pixar-style alternative (no reusable ID needed)

If you just want one cute character clip without continuity:

3D animated short, modern Pixar-quality rendering. A small spherical robot with one
large expressive eye rolls through a sunlit inventor's workshop. Wooden workbenches,
brass tools, dust motes in shafts of light. Camera tracks alongside the robot at
floor level. 10 seconds, 16:9, warm color grading.

Common failure modes

  • No reference image: text-only character descriptions drift quickly
  • Walk cycles: AI video universally struggles with locomotion — add “pauses and looks around” to give Omni a beat structure instead of pure walking
  • Hand interactions: holding cups, typing, gestures — fine details degrade. Frame to hide hands when possible
  • Crowd scenes: Omni’s character ID doesn’t scale to backgrounds with multiple faces

Sources