Glossary
Checkpoint
Short answer
The base file containing an image model's trained weights.
A checkpoint is the model's "brain": a large file holding all the weights produced by training. In the Stable Diffusion ecosystem, different checkpoints specialise in different aesthetics.
One checkpoint may excel at photorealistic portraits, another at anime, another at architectural visualisation. The same prompt on different checkpoints produces entirely different results.
Its difference from a LoRA is scale: a checkpoint is a complete model (gigabytes), while a LoRA is a small specialisation layered on top of one.