Flux.2 Breast Shape and Details
详情
下载文件 (1)
模型描述
This is my very first attempt at a LoRA so I made a simple one.
Trained with ostris/ai-toolkit on a single NVIDIA RTX Pro 6000 Blackwell (96GB VRAM).
Base model: FLUX.2-dev (black-forest-labs)
Method: LoRA, rank 32, alpha 32
Steps: 2000
Learning rate: 1e-4 (AdamW 8bit)
Batch size: 1
Optimizer: adamw8bit
Precision: bf16 with qfloat8 quantization
Gradient checkpointing: enabled
Timestep type: weighted
Noise scheduler: flowmatch
Dataset: 34 images at 512/768 resolution with text captions
Text encoder: Mistral Small 3.1 24B (frozen, embeddings cached to disk)
Trigger word:
flx2bdtEMA: enabled, decay 0.99
Caption dropout: 0.0



