Checkpoint
A checkpoint is a saved AI model file containing the full set of trained weights. It is the complete "brain" that generates images - swapping checkpoints changes the entire look and capability.
A checkpoint is a snapshot of a fully trained diffusion model - every weight it learned, saved to a file (commonly .safetensors). When a generator "uses a model," it is loading a checkpoint. It is the foundation everything else sits on: prompts, samplers and settings all operate through whichever checkpoint is loaded.
Checkpoint vs LoRA
A checkpoint is the whole model (large - often several gigabytes). A LoRA is a small add-on that adjusts a checkpoint without replacing it. You pick one checkpoint as your base, then optionally layer LoRAs on top for specific styles or characters.
Why it matters
The checkpoint is the single biggest factor in your output. Different checkpoints are tuned for different things - photorealism, anime, illustration, product shots - and each has its own ideal CFG scale and steps. PixelForge selects and tunes strong base models for you, so you get great results without managing model files yourself.
Try it in the generator
Put checkpoint to work right now - free daily generations, commercial license included.
Related terms
- LoRAA LoRA (Low-Rank Adaptation) is a small add-on file that teaches a base model a specific style, character, object or concept - without retraining the entire model.
- Diffusion modelA diffusion model is the type of AI that powers most modern image generators. It learns to turn random noise into a coherent image by reversing a step-by-step noising process.
- VAEA VAE (Variational Autoencoder) is the component that converts images between full-resolution pixels and the compressed latent space a diffusion model works in - encoding on the way in and decoding on the way out.
- CFG scaleCFG scale (classifier-free guidance scale) controls how strongly the image follows your prompt. Low values are loose and creative; high values stick closely to the prompt but can look over-processed.