app.py

import gradio as gr
from transformers import pipeline
from diffusers import StableDiffusionPipeline
import torch
import os

# 1. Check for HF_TOKEN
HF_TOKEN = os.getenv("HF_TOKEN")
if HF_TOKEN is None:
    raise ValueError("Please set the HF_TOKEN environment variable in Hugging Face repository secrets.")

# 2. Set device
device = "cuda" if torch.cuda.is_available() else "cpu"

# 3. Load translator with token
translator = pipeline(
    "translation",
    model="Helsinki-NLP/opus-mt-ta-en",
    use_auth_token=HF_TOKEN
)

# 4. Load text generator (GPT-2) — public, no token needed
generator = pipeline("text-generation", model="gpt2")

# 5. Load image generator (Stable Diffusion) with token
image_pipe = StableDiffusionPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    use_auth_token=HF_TOKEN,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32
)
image_pipe = image_pipe.to(device)

# 6. Main function
def generate_image_from_tamil(tamil_input):
    # Translate Tamil to English
    translated = translator(tamil_input, max_length=100)[0]['translation_text']

    # Generate a prompt using GPT-2
    generated = generator(translated, max_length=50, num_return_sequences=1)[0]['generated_text']
    generated = generated.strip()

    # Generate image using Stable Diffusion
    image = image_pipe(generated).images[0]

    return translated, generated, image

# 7. Gradio Interface
iface = gr.Interface(
    fn=generate_image_from_tamil,
    inputs=gr.Textbox(lines=2, label="Enter Tamil Text"),
    outputs=[
        gr.Textbox(label="Translated English Text"),
        gr.Textbox(label="Generated English Prompt"),
        gr.Image(label="Generated Image")
    ],
    title="Tamil to Image Generator",
    description="This app translates Tamil text to English, generates creative English prompts, and visualizes them using Stable Diffusion.",
    allow_flagging="never"
)

# 8. Launch app
iface.launch()