Define which base model to fine-tune (SDXL, Flux, SD3, etc.), training parameters like learning rate and batch size, LoRA/LyCORIS settings, and optimizer configurations.
Configure your training datasets, including image paths, caption handling, bucketing strategies, and preprocessing options. Each environment can have its own dataloader.
Create reusable collections of prompts for validation during training. Libraries can be shared across environments and used for consistent quality checks.
Define text processing rules for captions - remove unwanted phrases, replace terms, or apply regex patterns. Filters are applied during dataloader preprocessing.
Manage training configurations and caption filters
The Default environment provides base settings that are optionally inherited by other environments. When "Merge active environment defaults" is enabled in Basic Config → Project Settings, values from the Default environment are used as fallbacks for any settings not explicitly configured in the active environment.
| Name | Description | Type | Modified | Actions | ||
|---|---|---|---|---|---|---|
Environment:
|
() datasets webhook Algorithm: |
|
Create your first configuration or import an existing one.
arafed, ^remove this prefix, s/find/replace/