Plain-English explanation
An API is a contract, not a product a person looks at directly. It specifies exactly what a piece of software can ask for, in what format, and what it gets back, without either side needing to know how the other is internally built. Most people interact with AI APIs indirectly: they use an app with a chat window or a generate button, and that app's job is to translate the click into a properly formatted request to a model, then translate the response back into something to display.
When a company says it “offers an API,” it means developers can send requests directly, usually authenticated with an API key and billed per use, rather than only through that company's own consumer-facing app.
How it works
A typical flow: a client, an app, website, or script, sends a request to an endpoint address, including an API key and parameters such as the prompt, model choice, and image size, usually formatted as structured data. The provider's server runs inference on the requested model and returns a response, generated text, an image file or link, a video, back to the client in the same structured format. Rate limits, usage-based pricing, and content policy checks are typically enforced at this API layer, before a request even reaches the model itself.
Adult-AI use
A large share of adult AI companion apps and image or video tools aren't running their own models at all: they're a front end wrapping someone else's API, whether a mainstream LLM provider or a specialized image-generation service. Whatever content policy that underlying API enforces becomes a hard ceiling on what the app can produce, regardless of how permissive the app's own marketing sounds. Prompts or images submitted to the app typically pass through, and may be logged by, the underlying API provider as well. Local generation removes that provider layer, along with its policies and logging.
How it differs from local and hosted generation
API is about the interface, a defined way software talks to a model, not about where that model runs. Hosted generation describes running a model on someone else's infrastructure at all, which is usually accessed through an API, or a website built on top of one, but is the broader idea of “not your hardware.” Local generation typically involves no API call to another company at all in this sense, though local tools often expose their own API on your own machine so other local software can talk to them.
Common uses
- An app or developer calling a hosted image, video, or LLM provider's API directly rather than building a model from scratch.
- API keys required to connect a third-party front end to a provider's model.
- Usage-based API pricing, per token, per image, per second of video, distinct from a flat app subscription.
Limitations and misconceptions
- “API access” doesn't mean unrestricted access: almost all commercial AI APIs enforce their own content policy and rate limits regardless of what a downstream app promises its users.
- An app having “an API” isn't the same as an app being transparent about which underlying model it uses: many wrapper apps don't disclose this.
- Using an API still means data leaves your device to a third party's servers, the same basic privacy consideration as any other hosted-generation use.