1. Name Your Dataset: Create a simple English dataset name (e.g., "test") for your project. Avoid special characters and spaces to prevent technical issues during processing. This naming convention ensures compatibility throughout the training process.
2. Prepare Dataset: Upload a minimum of 10 high-quality images (1024x1024 resolution) with clean backgrounds. The higher quality and cleaner your source images, the better your LoRA model will perform. After upload, the system automatically generates necessary files - ensure all filenames remain in English characters to maintain compatibility.
3. Configure and Train: For model selection, you can either: (1) Use the pre-installed HunyuanVideo models (recommended for beginners) (2) Upload your custom HunyuanVideo model and VAE to the models directory (for advanced users) When using custom models, carefully modify only the model name while keeping the path structure intact. Click "Start Training" to begin the LoRA training process with your configured settings.
4. Monitor and Complete Training: Track the training progress through the outputlogs. Once you see "Saving model, Training complete" message, your LoRA model is ready. Navigate to File Storage > outputs to download your trained model for future use in HunyuanVideo generation."
