app.py

import os
import logging
import gradio as gr
from huggingface_hub import InferenceClient

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s | %(levelname)s | %(message)s",
    datefmt="%Y-%m-%d %H:%M:%S",
)
logger = logging.getLogger(__name__)

# Environment variables for configuration
HF_TOKEN = os.environ.get("HF_TOKEN", "")
CAPTION_MODEL = os.environ.get("CAPTION_MODEL", "Salesforce/blip-image-captioning-base")
VQA_MODEL = os.environ.get("VQA_MODEL", "dandelin/vilt-b32-finetuned-vqa")

logger.info(f"HF_TOKEN configured: {bool(HF_TOKEN)}")
logger.info(f"CAPTION_MODEL: {CAPTION_MODEL}")
logger.info(f"VQA_MODEL: {VQA_MODEL}")

client = InferenceClient(token=HF_TOKEN) if HF_TOKEN else InferenceClient()
logger.info("InferenceClient initialized")



def caption_image(image):
    """Generate a caption for the image."""
    logger.info(f"caption_image() called | image={image is not None}")
    
    if image is None:
        logger.warning("No image provided")
        return "📷 Upload an image first!"
    
    try:
        logger.info(f"Calling image_to_text | model={CAPTION_MODEL}")
        result = client.image_to_text(image, model=CAPTION_MODEL)
        logger.info(f"Caption: {result.generated_text[:100]}")
        return result.generated_text
    except Exception as e:
        logger.error(f"API error: {e}")
        return f"❌ Error: {e}"


def answer_question(image, question: str):
    """Answer a question about the image."""
    logger.info(f"answer_question() called | image={image is not None} | question={question[:30] if question else 'None'}")
    
    if image is None:
        logger.warning("No image provided")
        return "📷 Upload an image first!"
    if not question.strip():
        logger.warning("No question provided")
        return "❓ Ask a question!"
    
    try:
        logger.info(f"Calling visual_question_answering | model={VQA_MODEL}")
        result = client.visual_question_answering(image=image, question=question, model=VQA_MODEL)
        top = result[0]
        logger.info(f"Answer: {top.answer} ({top.score:.1%})")
        return f"🤖 {top.answer} (confidence: {top.score:.1%})"
    except Exception as e:
        logger.error(f"API error: {e}")
        return f"❌ Error: {e}"


logger.info("Building Gradio interface...")

with gr.Blocks(title="Vision Chat") as demo:
    gr.Markdown("# 👁️ Vision Chat\nUpload an image, get a caption, and ask questions about it!")

    with gr.Row(equal_height=True):
        with gr.Column(scale=1):
            img = gr.Image(type="pil", label="📷 Your Image")
            caption_btn = gr.Button("✨ Generate Caption", variant="primary")

        with gr.Column(scale=1):
            caption_out = gr.Textbox(label="Caption", lines=2, interactive=False)
            question = gr.Textbox(label="❓ Ask a question", placeholder="What color is the animal?")
            ask_btn = gr.Button("Ask", variant="secondary")
            answer_out = gr.Textbox(label="Answer", lines=2, interactive=False)

    caption_btn.click(caption_image, inputs=img, outputs=caption_out)
    ask_btn.click(answer_question, inputs=[img, question], outputs=answer_out)
    question.submit(answer_question, inputs=[img, question], outputs=answer_out)

demo.queue()
logger.info("Starting Gradio server...")
demo.launch()