🖼️ Reverse Engineering • Image to Prompt
Reverse Prompt Engineering Guide
How to deconstruct any viral photo, 3D render, or concept art into exact optical parameters and lighting tokens.
1 Midjourney /describe Protocol
Midjourney Discord / Web /describe [upload image] -> Select Option 1 or 2 for structural composition, Option 3 or 4 for artistic lighting.
PRO HACK: Take the best descriptive phrase from button 1 and the camera/lighting tag from button 3, then combine them.
2 GPT-4o & Claude 3.5 Sonnet Reverse Prompting
ChatGPT / Claude Web Upload reference image with prompt: "Deconstruct this image into a photorealistic Midjourney v6.1 prompt: 1. Subject & Pose, 2. Camera hardware, sensor & focal length, 3. Lighting direction & shadows, 4. Color grading palette, 5. Specific aspect ratio."
PRO HACK: Instruct the model to avoid generic adjectives ("hyper-realistic") and focus on camera and lens optics.
3 CLIP Interrogator ViT-L/14
Stable Diffusion / Flux Pass image through CLIP ViT-L/14 interrogation model to extract exact artist name tokens and medium tags.
PRO HACK: Useful for uncovering obscure 1970s illustrators or specific 3D render engine shaders.
📋 Universal Vision LLM Prompt Extractor Template
Works in GPT-4o & Claude 3.5Act as a world-class prompt engineer and cinematographer. Deconstruct this uploaded image into an exact Midjourney v6.1 prompt formula: 1. Core Subject & Silhouette 2. Camera Hardware, Sensor & Focal Length (e.g. Hasselblad 85mm, Leica 35mm) 3. Specific Lighting Setup (e.g. volumetric rim light, golden hour chiaroscuro) 4. Surface Texture & Atmospheric Effects (e.g. wet marble, mist, film grain) 5. Aspect Ratio & Parameter Flags (--ar, --v 6.1, --stylize) Output ONLY the final compiled prompt in a single code block.