Can it read a prompt out of a video?

Yes. Add an MP4 or MOV and a representative frame is pulled from the clip and described, which is the fastest way to match a look across a whole video series.

Which image models are supported?

Prompts can be written for Midjourney, Stable Diffusion, Flux and DALL·E. The wording and the parameter syntax change with the target so the prompt is usable as-is.

Do I need my own API key?

Yes. Analysis runs through your own provider key, which is why there is no per-image charge from us and no queue. The key stays in your browser unless you deliberately sync it to an account.

Will the generated image look exactly like my reference?

No, and it should not. A prompt captures the describable qualities of an image — subject, light, palette, style — not its exact pixels. Expect a close relative of your reference rather than a copy.

Tools / Image & video to prompt

Image & video to prompt generator

Upload JPG, PNG or WebP images — or MP4, MOV and WebM clips analysed frame by frame — and get production-ready AI prompts with matching negative prompts. Everything runs with your own Gemini or OpenAI key from the settings page.

Drop images or videos here or browse your files

JPG, PNG, WebP, MP4, MOV or WebM · up to 60 files per batch · originals never leave your device

0 file(s) · 0 prompt(s)

Nothing here yet. Upload a few images or videos to get started.

About the image to prompt generator

A reference image holds far more information than most people can put into words: lens character, lighting direction, colour grade, material finish, mood. This tool reads an image or a video frame and writes the prompt that would recreate it, so you can iterate on a look you already like instead of guessing at adjectives.

What a good generated prompt contains

The output is structured rather than poetic. It names the subject and the action, then the framing and camera perspective, then the lighting setup, then materials and textures, then colour palette and grade, and finally the rendering style. That order matters because most image models weight earlier tokens more heavily, so a prompt that opens with the subject reproduces the subject reliably.

You also get a negative prompt and a set of parameter suggestions, because a look is rarely just a description — aspect ratio and stylisation strength do as much work as the words.

How to use it

1. Add a reference
Upload a still, or a video file — a frame is extracted automatically so you can reverse-engineer the look of a clip as easily as a photo.
2. Pick the target model
Midjourney, Stable Diffusion, Flux and DALL·E respond to different phrasing and different parameter syntax. Choosing the target rewrites the prompt in the dialect that model actually understands.
3. Choose detail level and language
Short prompts give the model room to interpret; long ones lock the composition down. Output can be written in your own language when that is easier to edit.
4. Export
Copy a single prompt, or export the whole batch as CSV or TXT and feed it straight into a generation queue.

Using it responsibly

Describing an image you did not make and regenerating it is not a way around copyright, and stock marketplaces treat close derivatives of protected work as an infringement rather than a submission. Use your own frames, your own client work, or images you hold rights to, and use the prompt as a starting point you then push somewhere new.

If the result becomes a stock submission, declare it as AI-generated. The metadata embedder writes that declaration into the file so the disclosure travels with it.

Questions about this tool

Can it read a prompt out of a video?: Yes. Add an MP4 or MOV and a representative frame is pulled from the clip and described, which is the fastest way to match a look across a whole video series.
Which image models are supported?: Prompts can be written for Midjourney, Stable Diffusion, Flux and DALL·E. The wording and the parameter syntax change with the target so the prompt is usable as-is.
Do I need my own API key?: Yes. Analysis runs through your own provider key, which is why there is no per-image charge from us and no queue. The key stays in your browser unless you deliberately sync it to an account.
Will the generated image look exactly like my reference?: No, and it should not. A prompt captures the describable qualities of an image — subject, light, palette, style — not its exact pixels. Expect a close relative of your reference rather than a copy.