app.py

from huggingface_hub import InferenceClient
import os
import base64
import json

class LoRAInferenceWrapper:
    def __init__(self, model_id, token):
        # Initialize the InferenceClient
        self.client = InferenceClient(model_id, token=token)

    def load_lora_weights(self):
        # Define the path to the LoRA model
        lora_model_path = "./lora.model.pth"  # Update to the actual file name

        # Check if the file exists at the given path
        if os.path.exists(lora_model_path):
            print(f"Found LoRA model at: {lora_model_path}")
            with open(lora_model_path, 'rb') as f:
                return f.read()  # Load the file content
        else:
            raise FileNotFoundError(f"LoRA model not found at path: {lora_model_path}")

    def preprocess_lora_weights(self, lora_weights):
        # Preprocess the LoRA weights (e.g., Base64 encoding for JSON compatibility)
        return base64.b64encode(lora_weights).decode("utf-8")

    def generate_with_lora(self, prompt):
        # Load and preprocess the LoRA weights
        lora_weights = self.load_lora_weights()
        processed_lora = self.preprocess_lora_weights(lora_weights)

        # Combine the prompt and LoRA data as a single input
        extended_prompt = json.dumps({
            "prompt": prompt,
            "lora": processed_lora
        })

        # Generate the output using the InferenceClient
        result = self.client.text_to_image(prompt=extended_prompt)
        return result

# Example usage
model_id = "stabilityai/stable-diffusion-3.5-large"
token = "hf_YOUR_HF_API_TOKEN"  # Replace with your Hugging Face token

# Initialize the wrapper
lora_client = LoRAInferenceWrapper(model_id, token)

# Generate an image with the LoRA file applied
prompt = "The same woman, smiling at the beach."
try:
    result = lora_client.generate_with_lora(prompt)
    print("Generated image:", result)
except Exception as e:
    print("Error:", str(e))