yuliblu3 LoRA
A LoRA (Low-Rank Adaptation) model trained for the Higgsfield zimage base model.
Model Details
| Property | Value |
|---|---|
| Base Model | zimage (Higgsfield) |
| LoRA Rank | 256 (reduced from 512 via importance-based truncation) |
| Dtype | bfloat16 |
| Trigger Word | yuliblu3 |
| Training Steps | 3,000 (96 epochs) |
| Training Images | 31 |
| Training Software | ai-toolkit v0.7.24 |
Usage
Use the trigger word yuliblu3 in your prompt to activate the LoRA.
Training Images
The /training_images folder contains the 31 images used to train this LoRA.
Files
yuliblu3_train_r256.safetensors— The LoRA weights (rank 256, ~1.3GB)
Rank Reduction
The original LoRA was trained at rank 512 (2.5GB). It has been reduced to rank 256 (~1.3GB) using importance-based truncation, which preserves the most significant rank dimensions while halving the file size.
- Downloads last month
- 21