AI image generated from the “Shattered Stone Style Transfer” prompt
Result produced with GPT Image 2 · @Samann_ai

Shattered Stone Style Transfer

The prompt1986 chars
{
  "task": "image_to_image_style_transfer",
  "input_image": "{{USER_IMAGE}}",
  "prompt": "Create a hyper-real 3D studio composition that recreates the main subject from the provided image as a fragmented stone assemblage. The subject must be built from separate, clearly detached rock pieces with small visible gaps between shards (no pieces merging). Material look: fragmented slate + sandstone shards with chiseled edges, crisp fractures, visible stone grain, micro-scratches, and realistic roughness. Color palette: predominantly dark slate with subtle warm-ochre sandstone accents. Lighting: soft studio key light from top-left, gentle fill, subtle contact shadows under each shard, realistic ambient occlusion in crevices, clean reflections kept minimal. Background: minimal off-white seamless backdrop, no texture. Framing: centered, clean, straight-on, subject fully readable. Add a few tiny debris chips floating or resting near the base for depth. Preserve the subject’s identity, proportions, and recognizable silhouette from the input image while transforming all surfaces into stone fragments. Hyper-real, high detail, sharp focus, 8k render quality.",
  "negative_prompt": "text, typography, logo, watermark, signature, extra props, busy background, fog, heavy bloom, cartoon, illustration, lowpoly, plastic, metal, glossy paint, melted shapes, merged fragments, unreadable subject, blur, noise, low resolution, oversharpening halos, distorted face/body, extra limbs, deformed geometry",
  "output": {
    "aspect_ratio": "use_input_aspect_ratio",
    "background": "off_white",
    "camera": {
      "angle": "straight_on",
      "framing": "centered",
      "distance": "medium"
    }
  },
  "params": {
    "style_strength": 0.75,
    "identity_preservation": 0.9,
    "detail_level": "very_high",
    "lighting_preset": "soft_studio_top_left",
    "shadow_intensity": "subtle",
    "gap_visibility": "clear",
    "debris_chips": "few_tiny",
    "no_text": true
  }
}

Swap these placeholders

  • {USER_IMAGE

Replace each bracketed token with your own subject before generating — leaving them in place makes the model render the literal words.

Credit

Collected from EvoLinkAI/awesome-gpt-image-2-API-and-Prompts · originally shared by @Samann_ai.