Guide
How this works
Ten minutes here will save you a lot of credit. Nothing below is marketing — it is what actually changes the picture you get.
Getting started
- 1
Create an account. Email and a password, no card.
- 2
Add credit. The smallest top-up is 5 €, which is 33 images. Credit never expires.
- 3
Describe an image and press Generate. Ten seconds later it is in your library.
Writing a prompt
A prompt is a brief, not a wish. The model is very good at doing what you say and very bad at guessing what you meant. Four things do almost all the work:
- Subject
- What is in the frame, concretely. "A ginger cat", not "a pet".
- Setting
- Where it is. This is what stops the image floating in a void.
- Light
- Where the light comes from and what kind. This single line decides whether it reads as a photo or a drawing.
- Medium
- 35mm film, oil on canvas, 3D render, pencil. Say it or the model picks for you.
Weak: a beautiful cat
Strong: A ginger cat asleep on a radiator, winter light through a dusty window, shot on 50mm film, shallow depth of field
Words that do nothing
"Beautiful", "amazing", "high quality", "8K", "masterpiece". They add length without adding information. Replace each one with a fact — a light, a lens, a colour, a material.
Using your own photos
Drop images into the studio (or paste them straight from your clipboard) and the model works from them instead of from nothing. You can use up to ten in one generation.
Say what must stay
This is the mistake nearly everyone makes. If you only describe the change, the model feels free to change everything else too. Write the instruction in two halves:
Keep the subject, the pose and the face exactly as they are. Replace the background with a foggy pine forest at dawn, light coming from behind.
Faces
Two or three photos of the same person, sharp and well lit, give a much better likeness than one. Say explicitly that the face must be preserved.
Sizes and versions
Square for social, wide for banners and video thumbnails, tall for posters and phone wallpapers. Pick the shape you actually need — cropping a square down to a banner throws away half the image you paid for.
Versions makes several different images from the same prompt in one go. Four is the right number while you are still hunting for a direction; one is right once you know what you want and are refining it.
What it costs
0.15 € per image. Not per prompt, not per minute — per image that comes out. Four versions cost 0.60 €.
If a generation fails, the credit goes straight back to your wallet automatically. You are never charged for an image you did not receive. Every movement is listed in Wallet → Movements, so you can always see where the money went.
When the result is wrong
- Looks fake
- Add a light source and a lens. "Soft window light from the left, 50mm, shallow depth of field."
- Generic
- Constrain the palette to two or three colours. An unlimited palette is what makes AI images look like AI images.
- Face drifted
- More reference photos of that person, and say the face must be preserved.
- Changed the wrong thing
- Name what must stay untouched.
- Text is gibberish
- Expected. Generate the image without text and add the type yourself afterwards.
- Nearly right
- Change one thing and regenerate. Changing three things at once tells you nothing about which one helped.
Rules and limits
- Reference images: PNG, JPEG or WebP, up to 12 MB each, ten per generation.
- Three generations can run at once per account.
- Do not upload another person's face to generate images of them without their permission.
- What you make is yours, including for commercial use.
- Your uploads and results stay in your library. You can download anything at any time.
The ? button
Top right of the studio, next to the theme switch — or just press ? on your keyboard. Every control that could puzzle anyone lights up with a small marker; click one and it tells you exactly what it does, in plain words, right where it is.
That is deliberate: you should never have to come back to this page to find out what a button does.