Skip to main content
Training is in limited availability. Contact your fal representative to enable it for your account. The training skill and the training card appear only for accounts with access.
Training lets you teach a model your subject or style. You provide a dataset, and the agent runs a LoRA fine-tune on a base model. The result is a model you can use in later generations.

​
Start a run

Attach a ZIP of training images, up to 2 GB, and describe the goal:
The agent chooses the training settings for the base model. Step count and learning rate are set server side and cannot be changed in the chat. You can set a trigger phrase, which the agent uses verbatim.

​
The training card

A training card appears in the chat with:
  • A status badge and a progress bar with current and total steps
  • An estimated time to finish
  • A Training loss chart that updates as the run proceeds
  • Sample outputs from the model at checkpoints
  • A download action for the trained weights
The agent checks progress for you and reports when the run finishes, with a recommendation on whether the result looks ready to use.

​
Use the result

Once training completes, ask the agent to generate with the trained model and the trigger phrase. The model is available to your account like any other fal model.