Prompts for photo generation: how to write requests that actually work
A short prompt structure, common mistakes, and templates for portraits, studio shots, and social media.
A neural net doesn’t “guess your intent” — it follows the wording. The clearer the scene, light, and framing, the closer the result is to what you imagined.
Basic structure
- Who is in the frame (person, couple, pet).
- Where (interior, street, studio) and what light.
- Style (photorealism, fashion, cinematic).
- Framing: close-up, half-body portrait, full body.
Example: “photorealistic portrait, woman, 30s, warm indoor light, linen shirt, shallow depth of field, 85mm look”.
What usually breaks the result
- A request that’s too vague: “a beautiful photo”.
- Ten styles at once: cyberpunk + renaissance + anime.
- Contradictions: “night” and “harsh midday light”.
How it works here
In MagicBox you don’t have to write a prompt from scratch: pick a ready-made look in the Mini App or describe the scene in your own words. The face comes from your photo — you don’t need to “draw it in words”.
Start in the Telegram bot.
Частые вопросы
Do I need an English prompt?
For “photo from description”, English usually works better. For ready-made looks a prompt isn’t required — your own photo is enough.
Why does the face “drift”?
Most often the source is small, in profile, or in heavy shadow. You need a clear front-facing portrait in even light.
Читайте также
How to animate a photo with AI: a step-by-step guide
A short clip from a portrait: which sources work and which mistakes to avoid.
Creating an AI photoshoot: ready-made looks instead of a studio
How the looks catalog works: your face, a ready scene, a result without settings.
Which AI model to choose: photo, video, and special effects
A short guide to platform models: when you need speed vs face quality.
Попробуйте в боте
Готовые образы и генерация — в Telegram, без отдельной регистрации на сайте.