captcha.py

import os
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'

from transformers import VisionEncoderDecoderModel, TrOCRProcessor
from PIL import Image
import io
import base64

# Load model and processor
processor = TrOCRProcessor.from_pretrained("anuashok/ocr-captcha-v3", use_fast=True)
model = VisionEncoderDecoderModel.from_pretrained(
    "anuashok/ocr-captcha-v3")

def resolve_captcha(image_path):
    # Check if input is base64 string
    if isinstance(image_path, str) and image_path.startswith('data:image'):
        # Extract the base64 data after the comma
        base64_data = image_path.split(',')[1]
        # Decode base64 to bytes
        image_bytes = base64.b64decode(base64_data)
        # Create PIL Image from bytes
        image = Image.open(io.BytesIO(image_bytes)).convert("RGBA")
    else:
        # Handle as regular file path
        image = Image.open(image_path).convert("RGBA")
        
    background = Image.new("RGBA", image.size, (255, 255, 255))
    combined = Image.alpha_composite(background, image).convert("RGB")

    # Prepare image for the model
    pixel_values = processor(combined, return_tensors="pt").pixel_values

    # Generate text
    generated_ids = model.generate(pixel_values)
    generated_text = processor.batch_decode(
        generated_ids, skip_special_tokens=True)[0]
    
    return generated_text