AI Prompt Generator
Describe what you want in plain words and get back a professionally engineered prompt, written the way the AI you're actually using expects to be spoken to - a Midjourney prompt, a Claude prompt and a Stable Diffusion prompt are three genuinely different kinds of writing. Covers chat and reasoning models, image models, video models and photo-restoration prompts, with 20 ready-made templates and a short explanation of the technique behind each generated prompt.
1. Which AI is this prompt for?
The structure changes completely between them - this is not one prompt with a different label on it.
Chat & reasoning models
Role, context, constraints and a locked-down output format.Image models
Subject, style, lighting, composition, camera - and exclusions.Video models
One continuous shot, with the camera move written in.2. Describe what you want, in plain words
3. Fine-tune it
Your engineered prompt appears here
You will get the ready-to-paste prompt, any negative prompt or parameters the model needs, and a short explanation of why it is built that way.
Why the same idea needs a different prompt for each AI
Most free prompt tools reformat your sentence and hand it back. That is not prompt engineering. The models have genuinely different input expectations, and writing for the wrong one is the most common reason a prompt underperforms:
- Chat and reasoning models (ChatGPT, Claude, Gemini, DeepSeek, Grok) respond to a stated role, separated context, numbered constraints and an explicit output format. Claude in particular parses XML-style tags reliably, so a prompt generated for Claude AI is tagged rather than bulleted, and a prompt generator for DeepSeek leaves the reasoning path to the model instead of scripting it.
- Image models split further. An AI prompt generator for Midjourney has to produce one dense comma-separated phrase with the subject first and
--nofor exclusions. A free Stable Diffusion prompt generator has to emit weighted keyword tags plus a real, separate negative prompt. The best prompt for DALL-E 3 is the opposite again: full descriptive sentences, with exclusions phrased positively because negations tend to summon the thing you excluded. - Video models (Sora, Runway) want a single continuous shot with the camera movement written into the prompt, not a sequence of cuts.
- Photo editing is its own discipline: restoring an old photo needs an explicit preserve-and-do-not list, or the model quietly redraws the faces it was supposed to protect.
Every generated prompt comes with a short explanation of the techniques used on it, so this works as a professional prompt engineering tool you learn from rather than a black box you keep coming back to.
Frequently asked questions
Is this AI prompt generator really free?
Yes. It is a free ChatGPT prompt generator, AI art prompt builder and prompt writer for every model listed, with no account, no credits and no watermark. It runs on free AI models, which means it is occasionally slow when they are busy - if you see a warming-up message, press Try again a few seconds later.
Can it write an AI prompt for Midjourney or Stable Diffusion?
Yes, and they come out differently. As an AI prompt generator for Midjourney it writes one dense comma-separated phrase with the subject first, quality words last and exclusions moved into the --no flag, plus optional parameters like --ar 16:9 --style raw. As a free Stable Diffusion prompt generator it switches to weighted keyword tags and always returns a separate negative prompt covering the artifacts that specific image is prone to, which is the single highest-leverage field for fixing hands, text and watermarks.
How do I generate a prompt for Claude AI, Gemini or DeepSeek?
Pick the model at the top and the structure follows it. A prompt for Claude AI is built from XML-style tags with your data placed before the instructions and room to reason in a thinking block. A prompt generator for Gemini produces a persona, numbered requirements and a short skeleton of the expected answer. A prompt generator for DeepSeek keeps the goal and success criteria sharp but deliberately does not script the chain of thought, because reasoning models do better when the working is left to them.
Can it write a prompt to restore an old or damaged photo?
Yes - it is one of the built-in templates. Photo restoration prompts live or die on their preserve list, so the generated prompt names each defect to repair (creases, scratches, water staining, fading, a torn corner) and then explicitly forbids the model from re-drawing or beautifying faces, changing anyone's age or expression, adding or removing people and objects, re-cropping, or smoothing skin into a plastic look. Paste it into an image editing model such as ChatGPT or Gemini along with your scan.
Do I have to write the description myself?
No. There are 20 built-in templates across image generation, photo restoration, copywriting, coding, business, video, music and general chat - things like a Midjourney product photo, a LinkedIn post from a rough idea, a Python function with pytest tests, a cold outreach email and an old-photo restoration. Pick one, edit the description to match your situation, and generate. Nothing you type is stored.