What tokens are in a neural network
A token is the internal unit of payment for generations. One image, one video or one voiceover deducts a certain number of tokens from the balance. The heavier the work for the neural network, the more is deducted.
In short
- Tokens do not burn out at the end of a session, they stay on the balance until you spend them.
- Different neural networks cost differently: a simple image is cheaper than a second of video.
- Within one neural network the price also changes with resolution and duration.
- The exact price is always shown on the generate button before you run.
Why generations cost differently
Behind every generation there is computation on third-party GPUs, and the harder the task, the more it costs. Drawing a 1024-pixel image takes seconds of work. Assembling five seconds of video with sound takes minutes. The difference in price reflects exactly that, not how “premium” the model is.
| What we do | Price per generation |
|---|---|
| Image from text | from 2 tokens |
| Editing an uploaded photo | from 2 tokens |
| Text-to-speech voiceover | from 2 tokens |
| Answer from a text neural network | from 1 token |
| Video from a photo or from text | from 8 tokens |
| Video quality enhancement | from 8 tokens |
This is the lower bound for each direction: the most affordable model in its group. See specific numbers for all neural networks and the plans on the pricing page.
What raises the price within one neural network
- Resolution. 2K and 4K cost more than the regular 1K, since the image takes longer to compute. Some models have no high resolutions at all, and then there is no choice.
- Video and audio duration. The base price usually includes a few seconds, then there is a surcharge for each additional one.
- Length of the text for a voiceover. A short line and ten pages do not cost the same.
To avoid wasting tokens
The most common loss is iterating on the same prompt in a heavy model. Find the composition on a fast and cheap one, then repeat the final shot with the same prompt in the heavy one. The second most common is 4K where the image goes into a social media feed: nobody will see the difference, but more will be deducted.
“Not enough tokens”: what it means
The message means exactly one thing: the balance is lower than the price of the selected generation. No money is charged and nothing breaks. Your options:
- Take a lighter model in the same group, the result for the task is often no worse.
- Lower the resolution or duration, if the slider is available.
- Top up the balance on the pricing page.
Frequently asked questions
Do tokens expire at the end of the month?
No, they stay on the balance until they are spent.
Are there free tokens?
A starting credit is given on registration, enough to try generating an image and see how everything works. Regular work needs a plan.
Are tokens charged if a generation fails?
We do not charge for a technical failure on our side. But if the neural network did its work and produced a result you did not like, the generation counts as completed: the provider billed it as regular work.
Why did the same model charge a different amount?
It means the settings differed: resolution, duration or text length.