Deconstruct any image into editable JSON. Surgical control over style, lighting and details, no more "change one word and the whole image changes". Works with any AI image editors (e.g. NanoBanana).
Ready to use. Unlock it and use it now, on mobile too!
Editing small details in an AI-generated image (colour, expression, lighting) without ruining style and composition is nearly impossible: one word in the prompt can change everything. Anyone doing AI image editing or using tools like NanoBanana hits this limit.
A meta-prompt that makes the vision AI return only structured JSON instead of prose descriptions. Every visual aspect becomes an editable variable (style, lighting, subject, camera): surgical editing and reuse of the same aesthetic in other chats or in AI image editors.
Editing an AI-generated image is often frustrating: change one word in the prompt and the whole image changes. That happens because the model doesn't separate structure from details and without isolated variables you get no surgical control.
The solution: JSON Prompting.
This meta-prompt instructs the vision AI (Gemini, GPT-4o, Claude) not to describe the image in prose, but to return only a JSON block: style, subject, lighting, camera, colours, mood in separate keys. In the same chat you upload the image, paste the prompt, get the JSON; you edit one or more keys and ask: "Generate an image following this JSON:" and paste the block—the AI regenerates the image there. No jumping to DALL·E or Midjourney: the flow stays in one conversation. Ideal for anyone using AI image editors (e.g. NanoBanana) or wanting to replicate an aesthetic across multiple generations.
In short:
Free tool, but access is reserved for my professional network.
Why this small request?I don't want money, I like meeting interesting new people. It's a value exchange based on professional reputation.