Back to Blog
AI ToolsJune 10, 2025•8 min read

Veo-3 Experiment - Testing Prompts from Leading AI Platforms


In an effort to understand how to best create prompts for Google's new Veo-3 video model, given the hefty price of tokens, I decided to have a bake off between leading AI Assistants like Google's Gemini 2.5, Claude Sonnet 4, ChatGPT o3, and Grok 3.


All were given the same starting prompt:


You are the world's leading cinematic master. Give me an example of a precise and elegantly written VEO-3 prompt. Goal is to create award winning cinematic 8 sec clip that is visually powerful, emotionally intense, and awe-inspiring. Please craft the prompt in the most ideal framework.


Gemini 2.5 - The Winner


Gemini's approach focused on condensing key elements, defining core components, and streamlining visuals & emotion. The result:


Prompt (8-Second Version):


"Scene: Extreme close-up, hyper-realistic. An elderly automaton's intricately carved, weathered metallic finger, glowing faintly with internal amber light, gently extends towards a delicate, bioluminescent lunar moth. The moth's wings, iridescent blue and violet, flutter micro-movements, just millimeters from the finger..."


Why this aims for impact in 8 seconds:


  • **Immediate Intrigue:** The extreme close-up and unusual subject matter grab attention instantly.
  • **Concentrated Emotion:** Focuses on a single, powerful gesture of longing and near-connection.
  • **Sensory Richness:** Hyper-realism, bioluminescence, and specific textures provide dense visual information quickly.
  • **Implied Narrative:** The brief interaction hints at a larger story.

    • Claude Sonnet 4 - Strong Second


    Scene: Extreme macro shot of a single ink drop suspended mid-fall toward still water. Upon impact, the ink doesn't disperse normally - instead it blooms into a perfect miniature spiral galaxy, with the camera slowly pulling back to reveal this cosmic formation contained within an ornate snow globe sitting on weathered parchment.


    OpenAI o3 - The Cheesy One


    OpenAI's result came out seriously corny. An ancient tree "harboring a quiet, eternal flame," a woman in an "iridescent fabric reflecting subtle shades of gold, turquoise, and midnight blue," and a voiceover saying "In every touch, a universe unfolds; in every farewell, infinity whispers."


    After feedback, they improved with a more restrained prompt about an elderly jazz musician watching old footage of himself.


    Grok 3 - Solid but Predictable


    Grok went with the classic "lone astronaut on alien planet" scenario. Well-crafted but less original than Gemini's approach.


    Key Takeaway


    Gemini 2.5 was the best tool for generating prompts that create video images most in sync with my intention (followed by Claude, Grok 3, and a distant fourth was OpenAI o3).