app.py

import spaces
import os
import tempfile
import gradio as gr
from dotenv import load_dotenv
import torch
from scipy.io.wavfile import write
from diffusers import DiffusionPipeline
from transformers import pipeline
from pathlib import Path
from PIL import Image  # <-- Required for new model
import io               # <-- Required for new model

# --- Setup Models and Device ---

load_dotenv()
hf_token = os.getenv("HF_TKN")

# Use GPU if available, otherwise CPU
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

# Correctly initialize the modern, reliable captioning pipeline
captioning_pipeline = pipeline(
    "image-to-text",
    model="Salesforce/blip-image-captioning-large", 
    device=device
)
print("Image captioning pipeline loaded.")

# Initialize the audio pipeline. Use float16 for less VRAM on GPU.
pipe = DiffusionPipeline.from_pretrained(
    "cvssp/audioldm2",
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
)
print("Audio generation pipeline loaded.")


# --- Core Functions ---

@spaces.GPU(duration=120)
def analyze_image_with_free_model(image_file_bytes):
    """Takes image bytes and returns a caption."""
    try:
        print("Received image bytes, opening with Pillow...")
        # Open the image data directly from memory using Pillow
        image = Image.open(io.BytesIO(image_file_bytes)).convert("RGB")

        print("Generating caption...")
        results = captioning_pipeline(image)
        
        if not results or not isinstance(results, list):
            print("ERROR: Caption generation returned invalid results.")
            return "Error: Could not generate caption.", True
        
        caption = results[0].get("generated_text", "").strip()
        if not caption:
            print("ERROR: Generated caption is empty.")
            return "No caption was generated.", True

        print(f"Successfully generated caption: {caption}")
        return caption, False

    except Exception as e:
        print(f"!!!!!! EXCEPTION in analyze_image_with_free_model: {e}")
        return f"Error analyzing image: {e}", True

@spaces.GPU(duration=120)
def get_audioldm_from_caption(caption):
    """Takes a text caption and returns a filepath to a generated WAV file."""
    try:
        # Move the large audio pipeline to the GPU only when it's being used
        pipe.to(device) 
        
        print(f"Generating audio for prompt: '{caption}'")
        audio_output = pipe(
            prompt=caption,
            num_inference_steps=25, # Fewer steps for faster generation
            guidance_scale=7.0
        ).audios[0]

        # Move the pipeline back to CPU to free up GPU memory for others
        pipe.to("cpu")
        
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_wav:
            print(f"Saving audio to temporary file: {temp_wav.name}")
            # write(file, sample_rate, data)
            write(temp_wav.name, 16000, audio_output)
            return temp_wav.name

    except Exception as e:
        print(f"!!!!!! EXCEPTION in get_audioldm_from_caption: {e}")
        return None

# --- Gradio Interface ---

css = """
#col-container{ margin: 0 auto; max-width: 800px; }
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.HTML("""
    <h1 style="text-align: center;">🎶 Generate Sound Effects from Image</h1>
    <p style="text-align: center;">
        ⚡ Powered by <a href="https://bilsimaging.com" target="_blank">Bilsimaging</a>
    </p>
        """)

    gr.Markdown("""
    1. **Upload an image**.
    2. Click **Generate Description**.
    3. Click **Generate Sound Effect**.
    """)

    image_upload = gr.File(label="Upload Image", type="binary")
    generate_description_button = gr.Button("Generate Description", variant="primary")
    caption_display = gr.Textbox(label="Image Description", interactive=False)
    generate_sound_button = gr.Button("Generate Sound Effect")
    audio_output = gr.Audio(label="Generated Sound Effect")

    gr.Markdown("""
    ## 👥 Contribute & Support
    For support, questions, or to contribute, please contact us at 
    [contact@bilsimaging.com](mailto:contact@bilsimaging.com).
    Support our work and get involved by donating through 
    [Ko-fi](https://ko-fi.com/bilsimaging). - Bilel Aroua
    """)
    
    # --- Event Handlers ---

    def update_caption(image_bytes):
        """Wrapper function for the button click."""
        if image_bytes is None:
            return "Please upload an image first."
        description, _ = analyze_image_with_free_model(image_bytes)
        return description

    def generate_sound(description):
        """Wrapper function for the button click."""
        if not description or description.startswith("Error"):
            gr.Warning("Cannot generate sound without a valid description!")
            return None
        audio_path = get_audioldm_from_caption(description)
        if audio_path is None:
            gr.Error("Failed to generate audio. Please check the logs.")
        return audio_path

    generate_description_button.click(
        fn=update_caption,
        inputs=image_upload,
        outputs=caption_display
    )

    generate_sound_button.click(
        fn=generate_sound,
        inputs=caption_display,
        outputs=audio_output
    )
    
    gr.HTML('<a href="https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2FGenerate-Sound-Effects-from-Image"><img src="https://api.visitorbadge.io/api/visitors?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2FGenerate-Sound-Effects-from-Image&countColor=%23263759" /></a>')

# Launch the app. `share=True` is not needed on Spaces.
demo.launch()