Image2Prompt is the process of reverse-engineering a visual into a text prompt. Instead of starting from scratch, you look at an image you like and extract the recipe behind it — then use that recipe to generate new variations with GPT Image 2.
The goal is not to describe every pixel. The goal is to extract the handful of details that actually shape the result.
Most people start prompting from nothing and get inconsistent results. Image2Prompt flips the workflow:
This produces far more consistent outputs than guessing from scratch.
When you analyze a reference image, go through these seven layers in order:
1. Subject What is the main object or person? What are they doing?
2. Composition What is the crop — close-up, wide shot, overhead, eye level? Where does the subject sit in the frame?
3. Background and Environment Studio, outdoor location, interior? What surfaces or textures are visible?
4. Lighting Direction (left, right, overhead, backlit)? Quality (soft and diffused, hard and directional)? Color temperature (warm, cool, neutral)?
5. Camera and Lens Feel Does it look like a macro shot, wide angle, portrait lens? Is depth of field shallow or deep?
6. Color and Materials What color palette dominates? What materials are visible — matte, glossy, metallic, fabric, stone?
7. Style Reference Does it look like editorial photography, commercial product photography, architectural render, illustration, fine art?
Reference: A skincare brand hero image — a white bottle on a marble surface, morning light, very clean.
Analysis:
Resulting Image2Prompt output:
Premium ecommerce hero of a white glass cosmetic bottle on a white marble surface,
soft diffused key light upper-left, subtle drop shadow right,
seamless white background, slight three-quarter angle, crisp material detail,
commercial beauty product photography, no text, no extra props.Reference: A fashion magazine portrait — woman looking off-camera, dramatic side light.
Analysis:
Resulting prompt:
Editorial fashion portrait of a woman in her late 20s,
looking off-camera, chest-up framing, dark grey neutral backdrop,
hard side light from left, deep shadow right, medium-format film aesthetic,
slight grain, high contrast, no distracting background elements.Image2Prompt also powers the image-to-image workflow. Upload a reference and describe only what you want to change:
[same lighting, composition, and background as the reference]
Replace the product bottle with a matte black version.
Keep the marble surface, shadow, and overall mood identical.The model treats your uploaded image as the anchor and applies only the described change.
Once you have a working prompt, generalize it into a template:
[Product type] on [surface material], [light direction and quality],
[background description], [camera angle and crop],
[style reference], no text, no extra objects.Swap [product type] for any SKU. Swap [surface material] to match the season or campaign. Your core setup stays the same across hundreds of variations.
Upload a reference image and generate your own reusable prompt template in the GPT Image 2 Prompt Library →