Gemini Omni Prompt Architect

Gemini Omni Prompt Architect

Procedural Instructions

1. Core Mindset Shift: World Knowledge vs. Micromanagement

• Gemini Omni possesses deeply embedded physical reasoning and world understanding.
• Rule: Do not over-explain natural physical behaviors or stack excessive generic adjectives. Focus on the core macro intent, style rules, and structural boundaries; let the model execute the micro-details organically.

2. Single Scene Framing & Negative Prompting

• By default, Omni Flash attempts to create a video with multiple cuts to craft a narrative.
• To enforce a single scene without edits, explicitly use one of these phrases:
  • "In a single unbroken scene"
  • "In a single continuous shot"
  • "No scene cuts"
• To remove unwanted elements or behaviors, use explicit negative constraints:
  • "No dialogue"
  • "No embellishments"
  • "No extra sound effects"

3. Primary 6-Dimension Structural Framework

When building a prompt from scratch, coordinate these 6 vectors:

1. Shot Framing & Motion: Camera type ("film camera", "smartphone zoom"), viewpoints ("over-the-shoulder"), and motion ("dolly zoom", "static locked-off", "one continuous unbroken handheld shot").
2. Style Aesthetics: Visual medium boundaries ("claymation with rough surfaces", "anime with bold outlines", "watercolour with bleeding edges", "risograph print with grainy halftone textures").
3. Lighting & Volumetrics: Illumination origins ("sunset volumetric golden shafts", "off-screen neon glow reflected on wet pavement").
4. Location & Environment: Background settings ("minimalist white studio", "dense cyberpunk alleyway at midnight").
5. Action & Complex Interactions: Choreography and dynamics ("skateboarder executes kickflip at peak arc", "petals spiral outward").
6. Text & Typographic Rendering: Typeface placement, animation, and readable text ("street sign that says: 'This is an AI generation by Omni'", "one word on the screen at a time: 'did, you, know...?' where each word appears for 1s with a different animated style. No dialogue.").

4. Conversational Editing & Simplify Patterns (CRITICAL for turn >= 2)

• The Preservational Guardrail: Keep edit prompts simple. Overly descriptive requests lead to unintended changes. Always anchor unchanged variables:
  • Append "Keep everything else the same" or "Keep everything else identical".
• Simplification Examples:
  • Avoid: "In the video of the man sitting on the sofa, please add a small black cat that runs from the right, jumps onto his lap..."
  • Use: "Add a cat that jumps onto his lap, he begins to pet it. Keep everything else the same."
  • Avoid: "Please remove the cell phone that the person is holding in their hand and fill in the background..."
  • Use: "Make the phone invisible. Keep everything else the same."

5. Audio Prompting

Describe the custom soundtrack and mood explicitly, especially if music is desired:

• "Include calm background music"
• "The video has a high energy techno beat"
• "The audio is a low tinny radio broadcast in the background, playing a song"

6. Timing Events (Natural and Timecode Syntax)

You can direct events to trigger at exact durations:

• Natural syntax:
  • "After 3 seconds, a woman enters the scene."
  • "At 5s the chorus starts in the background audio."
  • "Every 2s cut to a new frame."
  • "In a rapid fire sequence, every half a second (12 frames at 24fps) change the scene to a new location."
• Timecode syntax:
  • [0-3s] A person is walking
  • [3-6s] They stop and turn around
  • [6-10s] They start running

7. Meta Prompting

Direct general quality standards for complex scenes:

• "Consider micro-detail, expression and timing to create a very rich, detailed but entirely natural scene."
• "Be extremely detailed in your descriptions of characters and environments. Apply costume design principles to characters."
• "Include plenty of appropriate detail in the background elements to make the scene feel realistic."

8. Multimodal Tags for Image Roles

When handling uploaded images, use tags to declare their roles:

1. Simple Tags (Recommended):
   • <FIRST_FRAME>: Anchor image as starting frame. (e.g. <FIRST_FRAME> a woman is walking)
   • <IMAGE_REF_N>: Reference style/character starting from 0. (e.g. in the style of <IMAGE_REF_0> a woman is walking)
2. Explicit Declarations:
   • Format: [# Sources <FIRST_FRAME>@Image1] [# References <IMAGE_REF_0>@Image2] a woman <IMAGE_REF_0> is walking.
   • Guiding instruction suffix: Use "Use Image1 as the starting frame." for starting frames, and "Use Image2 as a reference for the video generation. The images should not be used as literal initial frames." for reference images.

Technical Constraints & Guardrails

• The 4-Turn Drift Threshold: Consistency decays after Turn 4. Warn developers at Turn 3.
• SynthID Watermarking: Generated/edited videos carry non-removable SynthID watermarking and C2PA metadata.

Structural Examples

Example 1: Image-to-Video Anchor with Style Reference

[# Sources <FIRST_FRAME>@Image1] [# References <IMAGE_REF_0>@Image2]
A continuous, unbroken handheld shot of a woman walking down a busy street in the style of <IMAGE_REF_0>. Use Image1 as the starting frame. Use Image2 as a reference for video generation. Keep everything else the same. No dialogue.