A meter that only works inside one service
AI credits are a unit each service defines to measure use of features such as image generation, video generation, and AI agents. A monthly plan typically includes an allowance, and the balance falls each time you run one of those features.
Credits are not money. They are a meter that works only inside the service that issued them. One credit in Figma, Adobe Firefly, and Kimi does not represent the same value or the same consumption, and credits cannot be moved between services or compared on a shared scale.
What consumes credits
- Generating an image or a video
- Sending a request to a higher-capability model
- Handing an AI agent a task with several steps
- Expanding an image, removing a background, or upscaling
- Processing long documents or many reference files
Some features cost a fixed number per run. Others vary with the model you pick, the complexity of the job, how much text was processed, and how many files you attached. A short instruction can still consume a lot if the AI does a lot of work behind it.
How this differs from tokens and API pricing
Tokens are the small units a model uses to read and write text. API pricing is commonly calculated from input and output token counts.
AI credits repackage that complexity into a single consumer-facing number. Even when the underlying calculation uses token counts, image resolution, and model choice, there is rarely a fixed rate of one credit to a set number of tokens. Running out mid-month means waiting for the reset or buying more, even on a paid plan.
Spending fewer credits
- Check your balance, reset date, and overage settings before a large run
- Prototype with a lighter model or a lower resolution
- Decide the finished state and the changes you want before you prompt
- Revisit the source material or the instruction instead of retrying the same failure
- Confirm whether unused credits roll over
Credits spent and output quality are not proportional. Sending a vague instruction to an expensive model costs more than clarifying the goal and choosing a model that fits it. Credit costs and pricing change, so check the current figures in the service itself.