Workspace Map

• Top Header: collapse/expand training sidebar and close.
• Left Sidebar: model selector, training-data upload, trigger phrase, parameter controls, and start-training action.
• Main Panel: selected model details and training guidance.
• Bottom Jobs List: queue/progress tracking for all runs with download/cancel actions.

Panel-by-Panel Reference

Training Model Selector

Chooses the trainer backend. Available models are pulled from provider metadata with fallback defaults when provider tags are unavailable.

Training Data Upload Panel

Supports ZIP upload and, for supported trainers, multi-file selection workflows that are packaged for training. Includes drag-drop, size checks, validation, and preview feedback.

Training Data Helper Tools

Includes guide/help modal and project-aware data preparation flow for auto-packaging captioned training datasets.

Trigger Phrase Field

Sets the invocation phrase used later in generator prompts to activate trained LoRA behavior.

Core Hyperparameters Panel

Steps and learning rate controls. Ranges and defaults adapt to selected trainer type.

Model-Specific Parameters Panel

Conditional options shown per trainer, such as reference image count, caption defaults, output format, face detection/cropping, masks, style mode, synthetic captions, and auto-scale input for video trainers.

Start Training Action

Submit button is validation-aware and disabled when required data is missing or validation fails.

Step-by-Step Workflow

1. Prepare a focused, clean training set for one identity/style target.
2. Open LoRA Training and select the training model.
3. Upload training zip and resolve any validation errors.
4. Set trigger phrase and training parameters.
5. Click Start Training.
6. Track status in Training Jobs (queued, in progress, completed, failed).
7. Download completed LoRA outputs and apply them in generator prompts.

Training Jobs Panel

• Shows status icon, model id, progress bar, and relative creation time.
• Queued/running jobs can be canceled.
• Completed jobs expose direct download action for the trained LoRA.
• Failures show error messages inline for faster retry adjustment.

Data Quality Tips

• Keep references focused and consistent.
• Avoid mixed quality, mixed lens language, or conflicting style examples in one set.
• Version your datasets so successful training setups are reproducible.

Controls and Functions

Control	Function	Guardrail
Training data selector	Selects image assets or datasets for adapter training.	Use consistent, high-quality references for best results.
Subject/style fields	Define what the adapter should learn and how it should be triggered.	Keep identity, object, and style goals clear.
Training model picker	Selects the available LoRA training backend.	Model choice affects requirements and output compatibility.
Validation controls	Check dataset readiness before dispatch.	Fix missing or low-quality inputs before training.
Start training	Creates an async training task.	Monitor task state before using the adapter.
Import/register adapter	Adds a trained LoRA to the project or tool settings.	Confirm it appears in model/tool choices before production use.