Training Studio

Training Studio fine-tunes LoRAs on your own material, on your own GPU.

Project            base model type (character LoRA for video, or image LoRA)
├── Dataset        images with captions and a trigger word
│   └── Sample     one image plus its caption
└── Run            hyperparameters, live progress, status
    └── Checkpoint saved weights (intermediate and final)

Page	Covers
Datasets	Building a dataset from a character or from gallery images; crops, captions, the trigger word.
Train a character LoRA	The video-model LoRA: settings, expectations, progress, cancelling.
Train an image LoRA	The image-model LoRA.
Checkpoints and outputs	Where the weights go and how to use them.

One training run at a time: the GPU is shared with generation, and a run blocks the queue until it finishes or is cancelled.