app.py

import gradio as gr
import random

import requests
from PIL import Image
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
# from dotenv import load_dotenv


# Load the translation model
translation_model = AutoModelForSeq2SeqLM.from_pretrained("KarmaCST/nllb-200-distilled-600M-dz-to-en")
tokenizer = AutoTokenizer.from_pretrained("KarmaCST/nllb-200-distilled-600M-dz-to-en")


src_lang="dzo_Tibt"
tgt_lang="eng_Latn"

# def translate_dzongkha_to_english(text):
#     translation_pipeline = pipeline("translation",
#                                 model=translation_model,
#                                 tokenizer=tokenizer,
#                                 src_lang=src_lang,
#                                 tgt_lang=tgt_lang)
    
#     translated_text = translation_pipeline(text)[0]['translation_text']

#     return translated_text

model = gr.load("models/Purz/face-projection")

def generate_image(text, seed):
    translation_pipeline = pipeline("translation",
                                    model=translation_model,
                                    tokenizer=tokenizer,
                                    src_lang=src_lang,
                                    tgt_lang=tgt_lang)
    
    text = translation_pipeline(text)[0]['translation_text']

    if seed is not None:
        random.seed(seed)

    if text in [example[0] for example in examples]:
        print(f"Using example: {text}")

    return model(text)
examples=[
    ["བྱི་ཅུང་ཚུ་གངས་རི་གི་ཐོག་ཁར་འཕུར།", None],
    ["པཱ་རོ་ཁྲོམ་གྱི་ཐོག་ཁར་གནམ་གྲུ་འཕུར།",None],
    ["པཱ་རོ་ཁྲོམ་གྱི་ཐོག་ཁར་ ཤིང་ཚུ་གི་བར་ན་ གནམ་གྲུ་འཕུར་བའི་འཐོང་གནང་།",None],
    ["སློབ་ཕྲུག་ཚུ་ ཆརཔ་ནང་རྐང་རྩེད་རྩེ་དེས།",None]
]
# examples = [
#     ["Humanoid Cat Warrior, Full View", None],
#     ["Warhammer Sisterhood", None],
#     ["Future Robots war", None],
#     ["Fantasy dragon", None]
# ]

interface = gr.Interface(
    fn=generate_image,
    inputs=[
        gr.Textbox(label="Type here your imagination:", placeholder="Dzongkha text..."),
        gr.Slider(minimum=0, maximum=10000, step=1, label="Seed (optional)")
    ],
    outputs=gr.Image(label="Generated Image"),
    title="Dzongkha Text to Image Generation",
    examples=examples,
    theme="NoCrypt/miku",
    article="<h1>Created By:</h1>Mr. Karma Wangchuk<br>Lecturer<br>Information Technology Department<br>College of Science and Technology<br>Rinchending Phuentsholing<br>Chhukha Bhutan<br>",
    description="Sorry for the inconvenience. The model is currently running on the CPU, which might affect performance. We appreciate your understanding.",
)

interface.launch()