Pick a category, mix style + lighting + lens + mood + composition, and get a prompt formatted for any AI image model.
When to use it
Use the builder when you want consistent quality without memorizing every photographer's lighting term. The builder takes a structured approach: subject → style → lighting → lens → color → mood → composition, then formats it correctly for whichever AI model you're using. Midjourney expects comma-separated keywords plus flags; GPT Image-2 wants natural-language sentences; Stable Diffusion supports weighting syntax. The same selections produce a different output per model.
How to build a great prompt
- Pick your subject category. Portrait, landscape, product, logo, anime, cyberpunk, food, fashion or free. Each category presets the style and lens to sensible defaults.
- Describe one specific subject. Avoid lists of subjects. "young woman with curly red hair" beats "a person, maybe with hair". The clearer the noun, the more predictable the output.
- Mix one option per row. Each row (style, lighting, lens, color, mood, composition) gets one selection. More selections do not always help — models pick the strongest signal.
- Add 1-3 quality tags only. Stacking too many quality tags ("masterpiece, best quality, 8k, ultra HD, photorealistic, hyper detailed, intricate") confuses the model. Pick the 2 that matter most for your goal.
- Pick your model tab. Each model uses a different format. The output panel auto-formats the same selections for Midjourney, GPT Image-2, Flux 2, SD, and Nano Banana 2.
- Save what works. Hit Save current to keep good prompts in your browser. Reuse the structure and swap only the subject for a series of consistent images.
Common mistakes to avoid
- Switching subject mid-prompt — "a cat, no wait, a dog" — pick one.
- Mixing incompatible styles ("photorealistic anime") — models can't serve two masters.
- Copying a Midjourney prompt to GPT Image-2 verbatim — strip the --flags and rephrase as sentences.
- Forgetting negative prompts on Stable Diffusion / Flux 2 — they materially improve hands, faces, and anatomy.
- Saving 30 prompts you never reuse — keep only the structures that work for your style.
Frequently asked questions
How is this different from /tools/prompt-enhancer/?
Prompt Enhancer adds 4 elements (style/light/lens/mood) to a short idea. The Builder adds 7 elements (+ color, composition, quality tags, aspect ratio, negative prompt) and formats output per AI model. Use Enhancer for quick polish, Builder for studio-quality prompts.
Are my saved prompts private?
Yes. Everything is stored in your browser's localStorage. No data leaves your device, no account needed.
Does the builder generate images?
No — it generates the prompt text. Click the "try it →" link to open the recommended model and paste your prompt there.
Can I use one prompt across multiple models?
Yes, but switching tabs is the better workflow. The same selections become different prompts per model because Midjourney, GPT, Flux, SD, and Nano Banana each have their own syntax preferences.
What's the most important field?
The subject and the lighting. Style anchors the look, lighting carries most of the perceived quality. Get those right and everything else fine-tunes.