Public models: per output or per second
Popular official models such as the FLUX family carry a per-output price, shown on the model page. Most community models instead bill the seconds of hardware time each prediction uses, so the cost of one image depends on the GPU and on how long the model takes.