The AI prompt generator turns a rough request into a structured prompt that gets better answers from ChatGPT, Claude, Gemini and other models. Fill in the role, the task, the material, who it's for and the format you want back; rules and examples are added in clearly tagged sections. Switch to the image tab to build Midjourney, Stable Diffusion or DALL·E prompts with subject, style, lighting, camera and a negative prompt in each tool's own syntax. Everything happens in your browser — nothing is sent to an AI from this page.
| Tool | Aspect ratio | Leave things out |
|---|---|---|
| Midjourney | --ar 16:9 | --no text, people |
| Stable Diffusion / Flux | Set in the generator UI | Negative prompt field |
| DALL·E / ChatGPT | Described in words | "Avoid: …" in the prompt |
A clear task, the material to work on, who the answer is for, the format you want back and any rules. A role and an example of good output help further.
Tags such as <context> show the model where your instructions end and the material begins, which reduces it following instructions buried in pasted text.
Yes. The structure is model-neutral; paste the result into any chat model.
Midjourney reads parameters at the end, such as --ar 16:9 for the aspect ratio and --no for things to leave out. The generator adds them for you.
No. It only builds the prompt text in your browser for you to copy.
The AI prompt generator assembles the parts that make prompts work into a consistent order: a role line ("You are …", with a leading "Act as" or "You are" normalized so it isn't doubled), the task, the context or material, the audience, rules, examples and finally the output format. Tone becomes the first rule rather than a vague adjective. When XML tags are on, context, examples and rules are wrapped in <context>, <examples> and <rules> tags, which Anthropic and OpenAI both recommend so the model can tell instructions apart from the material it's working on. Rules are entered one per line and existing bullet markers are stripped and re-added, so pasted lists don't end up double-bulleted. Image prompts follow each tool's syntax: Midjourney gets a comma-separated description followed by --ar, --no and --stylize parameters; Stable Diffusion and Flux get the positive prompt and a separate negative prompt for the negative-prompt field; DALL·E and ChatGPT get a sentence-style prompt with "Avoid:" and the aspect described in words, because they don't accept parameters. Nothing is sent to any model; the output is text to paste.