What is a prompt rewriter?
A prompt rewriter takes a rough line and turns it into a prompt a specific model can follow. "A cat in a hat, cool" names a subject. It does not say how the picture is framed, what the light is, or which product you will paste it into. Those choices change the wording. Midjourney reads a sentence and then a few flags. Flux reads a caption. Stable Diffusion reads a positive prompt and a separate negative prompt. Nano Banana and GPT Image read a natural-language sentence. ChatGPT reads a task.
Reprompt (reprompt.org) is the prompt rewriter on this page. You paste the rough line, pick a target, and add a style or goal if you have one. You get one main prompt, two variations, and a short list of what changed. The tool is free and does not ask for an account. A pace limit and a daily ceiling keep it available. It writes the prompt. It does not generate the image, and it does not answer the ChatGPT task for you.
How do you rewrite a prompt?
- Paste the rough prompt in the box above. Short is fine. A few words is a normal starting point.
- Pick the target you will actually paste into. The same idea should not come back in the same shape for Midjourney and for ChatGPT.
- Add a style or goal if you care about one. "Cinematic, square crop" is enough. Leave it blank if you do not have a preference yet.
- Copy the main prompt. Read the notes before you swap in a variation. A variation is an alternate emphasis, not a second opinion you must use.
If you already have a picture and want the prompt pulled out of it, use image to prompt or the image to JSON prompt page. If you have a still and want a clip, use the image to video prompt tool. This page starts from words.
What does a real rewrite look like?
This is one run from October 6, 2026, using the same rewriter this page calls. The rough line was "a cat in a red hat, cool lighting, square". The style note was "cinematic, quiet". The target was Midjourney. The flags came back in their own field because the rough line said "square," which the model treated as a 1:1 frame.
Rough prompt
a cat in a red hat, cool lighting, square
Main prompt
A majestic cat wearing a stylish red hat, captured in a quiet, cinematic setting with soft moody lighting, intricate fur texture, shallow depth of field, high detail photography
Flags, separate from the sentence
--ar 1:1 --stylize 250
Two variations
A close-up portrait of a cat in a bright red beret, cinematic rim lighting, serene atmosphere, professional studio quality
A cat wearing a red fedora sitting in a dimly lit, cozy library, dust motes dancing in shafts of light, cinematic storytelling style
What changed
- Added descriptive adjectives to enhance visual detail and texture.
- Specified lighting and composition to align with the cinematic and quiet goal.
- Integrated photography terminology to ensure higher quality output.
The second variation moved the cat into a library. That is a bigger change than the first, which only tightens the portrait. If you wanted the original idea and nothing else, paste the main prompt and skip the library. The notes are the model's own account of the edit, kept here as they came back.
How should each model receive the prompt?
The rewriter changes the shape, not just the adjectives. A flag inside a Midjourney sentence is easy to delete by accident. A negative clause inside a Stable Diffusion positive prompt is in the wrong box. The table is the rule this page follows.
| Target | Paste this | Keep separate |
|---|---|---|
| Midjourney | The sentence, in /imagine or the web app. | --ar, --stylize, and any other flags, after the sentence. |
| Flux | The caption. Flux expects prose, not a weight syntax. | Nothing. Flags and a negative prompt are left empty. |
| Stable Diffusion | The positive prompt. | The negative prompt, in the negative field. |
| Nano Banana, GPT Image | The sentence, in the prompt box. Quoted words are words you want drawn. | Aspect ratio if the product has its own control. |
| ChatGPT text prompt | The task, with the constraint and the shape of the answer. | This is not an image prompt. No flags. |
When should you keep a variation?
Keep a variation when it changes one decision you were unsure about: a tighter crop, a different time of day, a calmer verb. Drop it when it changes the subject. The main prompt is the one to paste first. The notes exist so you can undo a change. If a note says the tool added a lens or a mood you do not want, delete that phrase. A rewritten prompt is a draft you can still edit.
A ChatGPT rewrite should get more specific about the job, not more purple. If you asked for a study plan, the rewrite should name the output, such as a week of tasks, and the constraint, such as 30 minutes a day. It should not turn the plan into a story. Image targets work the other way: they should add the light, the framing, and the material, because those are what the image model uses.
For a photo you already have, the prompt extractor is the better start. Rewrite that result here if you want it aimed at a different model. For Gemini image wording from scratch, the Nano Banana prompt generator follows that model's own order: subject, action, location, composition, style.
Prompt rewriter questions
What is a prompt rewriter?
A prompt rewriter turns a rough line into a prompt a specific model can follow. Reprompt (reprompt.org) takes that line, a target such as Midjourney, Flux, Stable Diffusion, Nano Banana, GPT Image, or ChatGPT, and an optional style note. It returns one main prompt, two variations, and a short list of what changed.
How is a prompt enhancer different from the text rewriter?
The prompt rewriter is for instructions you paste into an image model or into ChatGPT. The text rewriter at /text-rewriter edits a draft you wrote, for clarity and tone. Use this page when the output should be a prompt. Use the text rewriter when the output should be the finished paragraph.
Where do Midjourney flags go?
Midjourney flags such as --ar and --stylize are shown in their own box, after the sentence. Paste the sentence first, then the flags. Leaving them in the sentence makes them easier to lose when you edit a word.
Does Stable Diffusion get a negative prompt?
Yes. The positive prompt and the negative prompt are separate. Paste the negative prompt into the negative field in Automatic1111, Forge, or ComfyUI. It is not part of the positive sentence.
Can I paste the result into Nano Banana or GPT Image?
Yes. Pick Nano Banana (Gemini) or GPT Image as the target and paste the main prompt into that product's prompt box. This page does not generate the image. Flux gets a caption, with no flags and no negative prompt.
Do you store the prompt I paste?
The prompt is sent so it can be rewritten and is not stored as an account history. There is no account. A pace limit and a daily ceiling keep the free tool available. Sexual or explicit requests, anything involving a minor, and non-consensual edits of real people are refused.
Related tools: image to prompt, image to JSON prompt, image to video prompt, and the Nano Banana prompt generator, the JSON prompt generator, and the art style identifier.