|
|
--- |
|
|
license: artistic-2.0 |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- black-forest-labs/FLUX.1-dev |
|
|
tags: |
|
|
- miazelu |
|
|
--- |
|
|
Mia Zelu – LoRA for FLUX1 |
|
|
Trigger word: miazelu01 |
|
|
Base Model: FLUX1 Kontext / T5XXL |
|
|
|
|
|
Overview |
|
|
Mia Zelu is a highly consistent LoRA trained for the FLUX1 framework. She’s designed to represent a modern teen influencer with photogenic features ideal for fashion editorials, AI-driven social content, and stylized digital campaigns. Trained to maintain facial coherence across multiple lighting, outfit, and expression changes, this LoRA is perfect for both static portraits and dynamic use cases such as AI video generation or virtual try-on. |
|
|
|
|
|
Prompt Usage |
|
|
Always start your prompt with miazelu01 to lock in the character identity. |
|
|
|
|
|
Example Generation Prompt: |
|
|
miazelu01, full-face studio portrait, white background, soft shadows, natural makeup, fashion editorial style |
|
|
|
|
|
|
|
|
Technical Training Details |
|
|
Model: FLUX1 + T5XXL |
|
|
Trigger word: miazelu01 |
|
|
Training steps: 1000 |
|
|
Batch size: 1 |
|
|
Epochs: 10 |
|
|
Network Alpha: 128 |
|
|
Network Dim: 192 |
|
|
Precision: bf16 |
|
|
Optimizer: Adafactor |
|
|
Scheduler: constant |
|
|
Loss: L2 |
|
|
Learning Rates: |
|
|
UNet: 5e-5 |
|
|
Text Encoder: 5e-5 |
|
|
T5XXL: 5e-5 |
|
|
Gradient Checkpointing: Enabled |
|
|
Mixed Precision: bf16 |
|
|
Train T5XXL: True |
|
|
Clip Model: clip_l.safetensors |
|
|
Base Model Checkpoint: flux1-dev.safetensors |
|
|
Image Resolution: 1024×1024 |
|
|
Max Resolution: 1360×1360 |
|
|
Augmentations: None |
|
|
Cache Latents: True |
|
|
High VRAM Mode: Enabled |
|
|
|
|
|
License |
|
|
Open for creative, research, and commercial use. Credit to the creator is appreciated. |
|
|
Do not use in harmful, NSFW, or misleading contexts. |
