Checkpoint (base model)

A checkpoint is a complete, saved set of trained weights for an image model, essentially the file that contains everything the model has learned. Loading a different checkpoint changes the overall look and capabilities of your generations, since each one is trained on different data and can favor different styles, realism levels or subjects.

In tendre.red, SDXL checkpoints such as CyberRealisticXL, RealVisXL and DreamShaperXL are offered at first setup, with more (including JuggernautXL) available later from the Components tab. Pick one as your base model before generating, and any LoRA you use is applied on top of it.