railway-lora-sdxl

SDXL LoRA - Indian Railway Locomotives & Coaches

This is a Low-Rank Adaptation (LoRA) fine-tuned SDXL model specialized in generating high-quality, realistic images of Indian Railways — including locomotives, passenger coaches, and railway-related scenes.

📌 Model Details

• Base Model: Stable Diffusion XL (SDXL)
• Training Framework: Kohya Trainer
• Adaptation Method: LoRA
• Resolution: 1024x1024
• Precision: FP16
• Trained On: Custom-curated dataset of Indian Railway images (uploaded here)

🎯 Intended Use

• Generate images of:
  • Indian Rail locomotives (WAP-7, WDP-4D, etc.)
  • Passenger and freight coaches (LHB, ICF, SE1, etc.)
  • Railway scenes and stations
• Useful for railway enthusiasts, educational content creators, and transport visualization tools.

🧾 Example Prompt

Indian Railways WAP-7 locomotive hauling LHB coaches on a broad gauge track, realistic lighting, cinematic composition, 4k quality

🚫 Limitations

• Not suitable for generating non-railway content
• May require prompt tuning for fine results
• Might reflect visual biases from the training data

🛠 How to Use

from diffusers import StableDiffusionXLPipeline
from peft import PeftModel
import torch

# Load SDXL base
pipe = StableDiffusionXLPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")

# Load LoRA weights
pipe.unet.load_attn_procs("YOUR_USERNAME/YOUR_MODEL_NAME")

# Generate
prompt = "Indian Railways SE1 coach interior, ultra detailed, cinematic light"
image = pipe(prompt, num_inference_steps=28, guidance_scale=7).images[0]
image.save("output.png")

🧠 Training Details

• Epochs: 10
• Batch Size: 4
• Learning Rate: 1e-4
• Noise Offset: 0.0357
• Resolution: 1024x1024
• LoRA Config: dim=128, alpha=128

🧾 License

This model is released under the CreativeML Open RAIL-M license. Use responsibly and credit the author when applicable.