Train Custom LoRA Model
Fine-tune state-of-the-art architectures on your own characters, subjects, or artistic styles.
Choose Trainer Model Architecture
FLUX.1 [dev] Subject
Fine-tune custom characters, people, products, and objects with state-of-the-art FLUX.1 [dev] fidelity.
FLUX.1 [dev] Style
Specialized for learning artistic styles, illustration aesthetics, color palettes, and distinct visual themes.
FLUX.2 [klein] 4B Style
Lightweight and ultra-fast 4B architecture for rapid style adaptation and responsive generation.
FLUX.2 [klein] 9B Style
Deep 9B parameter architecture delivering rich details, complex textures, and high-fidelity style preservation.
Qwen-Image LoRA
Accelerated fine-tuning for Qwen-Image models for strong prompt alignment, character, and concept consistency.
Qwen-Image 2512 LoRA
Advanced checkpoint offering superior photorealism, fine human skin textures, typography, and complex scenes.
Z-Image LoRA
Auto-tuned hyperparameter training for Z-Image architecture with optimized rank and step scaling.
Z-Image Base LoRA
High-capacity fine-tuning on foundational Z-Image base architecture for rich photorealism and deep concept retention.
Training Identity & Trigger Word
Unique name used to identify this LoRA in your library.
The specific keyword you'll include in prompts to activate this trained concept.
Training Parameters & Settings
images_data_url•Trigger Field: trigger_phraseSpecifies whether the adapter optimizes for subjects or overall visual styles.
Upload Training Dataset
Dataset Guidance: Upload 10 to 50 clear, high-resolution images. Ensure varied angles, lighting, expressions, and backgrounds for subject models, or consistent aesthetics for style models.
0 images selected • 640 Credits