Wiro - Train Dreambooth Sdxl icon

Wiro - Train Dreambooth Sdxl

Train your own custom Stable Diffusion XL model using a small set of images

Wiro - Train Dreambooth Sdxl

AI-generated

Overview

This node enables training a custom Stable Diffusion XL model using a small set of images via the Wiro API. It is useful for users who want to create personalized AI models tailored to specific datasets and prompts, such as generating unique image styles or enhancing model capabilities with custom data. The node sends detailed training parameters to the Wiro API, initiates the training process, and polls for completion, returning the training status and result URL.

Use Case Examples

  1. Training a custom image generation model with specific prompts and dataset.
  2. Automating model training with custom learning rates, batch sizes, and precision settings.

Properties

Name Meaning
Trained Model Name The name of the model to be trained, required for identification.
Train Model Description Optional description of the training model.
Reference Model Reference model to base the training on.
Reference Model Private Privacy setting for the reference model.
Classification Dataset Dataset used for classification during training, required.
Instance Prompt Prompt describing the instance for training, required.
Class Prompt Prompt describing the class for training, required.
Save Sample Prompt Prompt used for saving sample outputs.
Learning Rate Controls the speed of model updates during training, required.
Resolution Image resolution used during training, required.
Seed Seed value for reproducibility, required.
Max Train Steps Maximum number of training steps, required.
Save Every N Steps Interval of steps to save the model, required.
Gradient Checkpointing Enable or disable gradient checkpointing for memory optimization.
Train Text Encoder Whether to train the text encoder component.
Full Bf16 Enable full bf16 precision for efficient memory usage.
Full Fp16 Enable full fp16 precision for efficient memory usage.
Train Batch Size Batch size of the training data, required.
Sample Sampler Sampler algorithm used for samples, required.
Cache Latents Enable caching of latent space representations to speed up training.
Cache Latents To Disk Save latent data to disk to reduce RAM usage; requires cache latents enabled.
Mem Eff Attn Enable memory efficient attention.
Xformers Utilize the xformers library for memory optimization and speed.
Use 8bit Adam Use 8-bit Adam optimizer for better memory efficiency.
Enable Bucket Enable buckets for multi aspect ratio training.
Lr Scheduler Learning rate scheduler type, required.
Lr Scheduler Num Cycles Number of cycles for the learning rate schedule, required.
Gradient Accumulation Steps Accumulates gradients over multiple steps to simulate larger batch size, required.
Max Grad Norm Limits gradient values to prevent explosion.
Mixed Precision Precision used during training, required.
Save Precision Precision level for saving the model, required.
Save Every N Epochs Interval of epochs to save the model, required.
Optimizer Type Optimizer algorithm used for training, required.
Optimizer Args Additional optimizer arguments such as weight_decay and betas.
Bucket Reso Steps Step size for resolution bucketing, required.
Noise Offset Noise added during training to improve generalization, required.
Bucket No Upscale Disable upscaling of images during bucketing.
Sample Every N Epochs Generates sample outputs every specified number of epochs, required.
Sample Every N Steps Generates sample outputs every specified number of steps, required.
Num Of Repeats Number of times each training image is repeated during training, required.
Cache Info Enable caching of meta information for faster dataset loading.
Output Config Output command line arguments to a .toml file.
Persistent Data Loader Workers Keep data loaders persistent for better performance.
Save State On Train End Save the training state at the end of the training process.
Selected Folder For Train Data Folder path for training data.
ZIP URL For Training Data URL to a ZIP file containing training data, automatically extracted.

Output

JSON

  • taskid - Unique identifier for the training task.
  • url - URL to access the trained model or results.
  • status - Status of the training task, e.g., 'completed' or 'failed'.

Dependencies

  • Wiro API with API key and secret for authentication

Troubleshooting

  • Error if the Wiro API does not return a valid task ID or socket access token, indicating an issue with the API request or credentials.
  • Ensure all required parameters are provided and valid to avoid API errors.
  • Network issues or API downtime may cause polling to fail or timeout.

Links

Discussion