PromptLens

AI images already carry their prompt. This reads it out — it does not guess.

Stable Diffusion, ComfyUI and NovelAI write the positive prompt, negative prompt, seed, sampler and model into the file itself. When there is nothing embedded, this says so plainly and falls back to an estimate from the picture — clearly labelled as an estimate. Everything runs in your browser; the image is never uploaded.

Drop an image here, or click to choose JPG / PNG / WebP / GIF · you can also just paste with Ctrl+V

What it reads, and what it cannot

Reads: aspect ratio, lighting key, saturation, dominant palette, detail density, compositional weight, resolution — measurable facts that map directly to lighting, tone and composition terms.
Cannot read: what objects or people are in the frame, or which art movement it belongs to. That needs a vision model, which means uploading your image — the whole premise here is that nothing leaves your device. For object-level description, a multimodal model is the right tool.
Best used as: technical terms appended to a subject description you write yourself. You know the subject; let measurement handle lighting and composition.

Why nothing is uploaded

Most reverse-prompt tools upload your image to their servers and call a model. Three consequences: your image sits on someone else’s machine, you get credit limits, and you wait in a queue. Running the analysis in the browser removes all three.

Never leaves your device

There is no upload endpoint; it works offline. Safe for client assets, unreleased work and private photos.

No credits

No API call means no cost, so there is no usage cap and no signup.

Reproducible

The same image always yields the same terms — it is measurement, not a model guessing.