Emily

Veo3 fast on the Gemini app 😀
I used the simple prompt from Artificial Analysis and used Kimi K2 to rewrite the prompt in JSON style, check the results, that's why natural language is not the best practice, each model will translate the language based on the text encoder and the data caption used for the training, the test should be writing the prompt in the best practice for each model, all these leaderboard are misleading and wrong. 

{
  "duration": "8s",

  "subject": {
    "primary": "vast, empty high-plains landscape",
    "character_attributes": "cracked clay pan, sparse tufts of dry grass, distant low mesas on the horizon"
  },

  "timeline": [
    {
      "start": "0.0s",
      "end": "3.0s",
      "camera": "locked-off ultra-wide shot, 1 m above ground, lens centre aligned with horizon (that's where the camera is)",
      "action": "first fork of lightning arcs left-to-right across storm front at 1.2 s; brief 0.3 s illumination reveals texture of cracked earth"
    },
    {
      "start": "3.0s",
      "end": "6.0s",
      "camera": "10 cm slow push-in, same height; faint handheld micro-shake for wind buffeting",
      "action": "multiple cloud-to-ground bolts strike at random intervals (3.4 s, 4.8 s, 5.5 s); each flash casts stark shadows and fades to near-black"
    },
    {
      "start": "6.0s",
      "end": "8.0s",
      "camera": "slight tilt up (3°) to capture towering cumulonimbus anvil; lens stays ultrawide",
      "action": "massive central lightning fork at 6.7 s—white-hot core, violet corona; ground shakes subtly as thunder climax rolls"
    }
  ],

  "scene": {
    "setting": "nighttime storm over flat desert plain—low humidity haze, ragged cloud base lit from within by sheet lightning",
    "physics": "realistic light-travel delay: thunder rumbles 0.8 s after each flash; dust plumes lift briefly under electromagnetic shockwave"
  },

  "style": "hyper-real 8K clarity, high dynamic range, cool blue-grey palette with white lightning cores, cinematic contrast",

  "audio": {
    "track": "minimal—sub-bass drone at −16 LUFS for tension",
    "sfx": [
      "rolling thunder: 1.9 s, 4.2 s, 6.0 s, 7.5 s (deep, spatial, 3-second decay)",
      "distant wind gusts, low howl, subtle grit of sand skittering past mic"
    ]
  },

  "technical": {
    "aspect_ratio": "16:9",
    "frame_rate": "30 fps",
    "lens": "14 mm rectilinear, f/4, infinity focus",
    "lighting": "natural—scene lit only by lightning; sensor pushes ISO during dark intervals"
  },

  "negative_prompt": "no text overlays, no watermarks, no subtitles, no unnatural lens flares, no inaccurate lightning colour, no low-resolution noise"
}