Image to JSON prompt converter

Why describe an image as JSON

A paragraph prompt mixes everything together; when you change one detail you often break another. A JSON prompt separates the decisions, which has practical advantages:

• Controlled variations: keep lighting and palette fixed and swap only subject to get a matching set of images.
• Reuse: store one JSON as a template for a brand or a series.
• Automation: scripts and spreadsheets can fill keys programmatically.
• Clarity for chat-based models: models such as Nano Banana or ChatGPT image generation read structured instructions well when you paste the JSON with a line like “Create an image from this description”.

JSON is not magic, though. Image models do not run code; they read the text. The benefit is organization and consistency, not secret extra quality.

The keys you get in each mode

When AI mode is available on this site, a vision model returns a JSON object with these keys: subject, setting, composition, camera, lighting, colors (an array of names), style, mood, details (an array), text_in_image, aspect_ratio and negative. Values are written in English, and real people are not identified by name. On this page the AI always answers in JSON, whichever target is selected.

Without AI mode, the analysis runs locally in your browser and returns a smaller object built from measurable pixel data. The subject is left in brackets for you:

{
  "subject": "[describe the main subject in a few words]",
  "lighting": "bright, high-key lighting",
  "palette": "pale yellow, white, teal",
  "look": "natural colors, soft, low contrast",
  "framing": "portrait",
  "aspect_ratio": "4:5"
}

The local mode cannot recognize objects, setting or style, so add those keys yourself if you need them, for example "style": "watercolor illustration".

A template workflow for consistent series

1. Upload the image whose look you want to match, for example your best product photo.
2. Generate the JSON and fill in or correct the subject.
3. Save it as your template. For every new item, change only subject and, if needed, details.
4. Paste each version into your image model with a short instruction, or convert it to plain text with the builder.

Edited example for an online shop:

{
  "subject": "a linen tote bag hanging on a wooden peg",
  "lighting": "bright, high-key lighting",
  "palette": "pale yellow, white, teal",
  "look": "natural colors, soft, low contrast",
  "framing": "portrait",
  "aspect_ratio": "4:5",
  "negative": "text, watermark, people"
}

Checking and using your JSON

Before pasting, make sure the JSON is still valid after editing: every key and string in double quotes, commas between fields and none after the last one. If a tool rejects it, the missing comma is usually the cause. To count how much space a long JSON prompt takes in a chat model, use the token counter. For a normal text prompt from the same picture, use image to prompt, and to write JSON from scratch, tick “Also give me a JSON version” in the image prompt generator.