How to use a neural network: your first run
Four steps from an empty screen to a finished image. Nothing to install, everything happens in the browser.
The order of steps
- Choose what you are making: an image, a video or a voiceover.
- Choose a neural network. They are grouped by name.
- Describe the result in words, that is the prompt.
- Check the price on the button and run.
Step 1. What you want to get
Neural networks are not universal: one draws images, another makes video, a third reads text aloud. Start with the task, not with the model name, so you land in the right section right away and do not choose from a hundred options.
Step 2. Which neural network to take
For the first run, take any of the inexpensive ones, the difference between models only becomes clear once you have something to compare with. An expensive model will not save a vague request, and a cheap one often gives the same result on a simple task.
A common beginner mistake
Starting with the most expensive model “so that it turns out well”. On the first runs the problem is almost always the description, not the model: an unclear prompt gives an unclear result in any of them. A cheap model is more useful at this stage, you will learn faster how to phrase things.
Step 3. Describe the result
A prompt is a description of what should come out. A simple scheme works: object → details → light → background → format.
A matte white ceramic mug on a wooden table.
Close-up, the object takes up two thirds of the frame.
Soft diffused light from the left, a light shadow on the right.
Background: a blurred kitchen in light tones.Read more about what a good request is made of in the card about prompts.
Step 4. Look at the price and run
The cost is shown on the generate button before you run and changes along with the settings: raise the resolution or duration and the number goes up. It is a convenient way to see what you pay for before you spend anything. How spending works is covered in the card about tokens.
What to do if you did not like the result
- The wrong object: name it with the very first word of the prompt, the model gives the beginning more weight.
- The wrong atmosphere: add light and background, that is usually the issue.
- It comes out different every time, that is how generation works. To make shots alike, describe the light, background and angle with exactly the same words.
- Completely off: try another model. They understand the same wording differently.