📌 LoRA Model – influence_girl
📝 Model Description (ENG)
This LoRA model is designed to generate hyper-realistic images of women in diverse photography styles, from casual social media selfies to high-end beauty and glamour portraits.
🎨 Multi-Style Flexibility! Seamlessly blend different aesthetics by adjusting the weight of trigger words.
Description
Learning Rates:
unetLR: 0.00035textEncoderLR: 0.00015
Clip Skip: 2 (helps avoid over-saturation or repetition in certain text encodings)
LoRA Settings:
loraType: “lora”networkDim: 32networkAlpha: 16 (maintains quality with reasonable VRAM usage)
Training Strategy:
numRepeats: 4keepTokens: 1shuffleCaption: true (improves generalization)flipAugmentation: false (avoids excessive mirroring)
Resolution: (Specify your desired pixel resolution, e.g., 512×512 or 768×768)
Scheduler:
cosine_with_restarts, withlrSchedulerNumCycles: 3Noise & Regularization:
minSnrGamma: 5noiseOffset: 0.15
Target Steps: 5400
Max Epochs: 12
Training Batch Size: 2