Checkpoint
A checkpoint is a saved file containing all of an AI image model's trained weights — the complete base model that a generator loads to turn text prompts into images.
June 16, 2026
A checkpoint is a saved snapshot of a fully trained AI image model's weights — the complete file the generator loads to turn prompts into images.
How it works
Training an image model adjusts millions or billions of numerical weights until the network can map text descriptions to pictures. A checkpoint is that trained state written to disk: it stores every weight needed to run the model. When a generator "loads a checkpoint," it reads this file into memory so the model is ready to produce images. The term comes from machine learning, where progress is periodically saved as a checkpoint during long training runs.
Because the checkpoint contains the entire model, it is large — frequently several gigabytes. It defines the baseline behavior of generation: the kinds of subjects the model renders well, its default aesthetic, its understanding of prompts, and its overall quality. Add-ons like LoRAs and ControlNets are layered on top of a checkpoint rather than replacing it.
Why it matters
The checkpoint is the foundation every other setting builds on. Two generators using different checkpoints can interpret the same prompt very differently — one tuned for photorealism, another for anime or illustration. Choosing or upgrading the checkpoint is the single biggest factor in baseline quality and style, which is why new model releases are such significant events.
In eaxy
eaxy runs and updates the underlying models so you never touch a checkpoint file. You choose a style pack and a prompt; eaxy loads and maintains the right base model behind the scenes, keeping quality high as newer models ship.
Related terms
Preguntas frecuentes
What is a checkpoint in AI image generation?+
It is the full set of a model's learned weights saved as a file. Loading a checkpoint loads the entire model, which determines the baseline style, quality, and capabilities of everything you generate.
What is the difference between a checkpoint and a LoRA?+
A checkpoint is the whole model, often several gigabytes. A LoRA is a small add-on layered on top of a checkpoint to push it toward one specific style or concept.
Why are there different checkpoints?+
Different checkpoints are trained or tuned for different goals — photorealism, anime, illustration, or general use — so each one gives a distinct baseline look and strength.
Do I choose a checkpoint in eaxy?+
No. eaxy runs and maintains the underlying models for you, so you simply pick a style pack and generate without managing checkpoint files.
Hazlo con eaxy
Describe cualquier cosa y genera imagenes increibles en segundos; despues dales movimiento con Kling 3.