Inference
Inference is the process by which a trained machine learning model produces an output from a given input, as opposed to training, which is the earlier process of adjusting the model's parameters using large datasets. When a user submits a text prompt or reference image to a generation tool, the resulting image or video is created through inference: the already-trained model applies its learned parameters to generate a new result without modifying itself.
On AI content generation platforms, inference is the step that consumes computing resources every time a user requests an output, which is why platforms often meter usage through credits, generation limits, or queue priority tied to subscription tiers. The speed and quality of inference depend on the hardware used, the size of the model, and settings such as resolution or number of generation steps, all of which a provider can adjust to balance cost against output fidelity.
For evaluating a platform, inference-related factors include how many generations are included in a plan, whether inference runs on the provider's own servers or through a third-party API, and whether images or videos are produced instantly or after a wait. Because inference is separate from training, a platform's inference behavior does not by itself reveal what data the underlying model was trained on; it only describes how outputs are generated once a model already exists.
Updated: