app.py

import gradio as gr
import torch
from transformers import (
    AutoProcessor,
    BlipForConditionalGeneration,
    pipeline,
    SpeechT5Processor,
    SpeechT5ForTextToSpeech,
    SpeechT5HifiGan
)
from PIL import Image

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

# ---------------------------------------------------------
# 1) IMAGE → CAPTION (BLIP)
# ---------------------------------------------------------
caption_model_name = "Salesforce/blip-image-captioning-base"
caption_processor = AutoProcessor.from_pretrained(caption_model_name)
caption_model = BlipForConditionalGeneration.from_pretrained(caption_model_name).to(device)

def generate_caption(image: Image.Image) -> str:
    inputs = caption_processor(images=image, return_tensors="pt").to(device)
    with torch.no_grad():
        output_ids = caption_model.generate(**inputs, max_length=30)
    caption = caption_processor.decode(output_ids[0], skip_special_tokens=True)
    return caption

# ---------------------------------------------------------
# 2) CAPTION → FAIRY TALE (Flan-T5)
# ---------------------------------------------------------
# Используем flan-t5-base или flan-t5-large (если есть память)
story_model = pipeline(
    "text2text-generation",
    model="google/flan-t5-base",
    max_new_tokens=180,
    device=0 if device == "cuda" else -1,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32
)

def generate_fairy_tale(caption: str) -> str:
    prompt = (
        "You are a kind storyteller for young children. "
        "Based on the following description, create a short, gentle, and imaginative fairy tale (3–4 sentences):\n\n"
        f"Image description: {caption}\n\n"
        "Fairy tale:"
    )
    result = story_model(
        prompt,
        temperature=0.9,
        top_p=0.92,
        do_sample=True
    )[0]["generated_text"]
    return result.strip()

# ---------------------------------------------------------
# 3) FAIRY TALE → SPEECH (SpeechT5 + HiFi-GAN)
# ---------------------------------------------------------
tts_processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
tts_model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)

# Используем фиксированный speaker embedding для стабильности
# (можно загрузить из датасета, но для демо — random с фиксированным seed)
torch.manual_seed(42)
speaker_embedding = torch.randn(1, 512).to(device)

def text_to_speech(text: str):
    # Ограничим длину, чтобы избежать переполнения
    text = text[:200]
    inputs = tts_processor(text=text, return_tensors="pt").to(device)
    with torch.no_grad():
        speech = tts_model.generate_speech(
            inputs["input_ids"],
            speaker_embedding,
            vocoder=vocoder
        )
    audio = speech.cpu().numpy()
    sample_rate = 16000
    return (sample_rate, audio)

# ---------------------------------------------------------
# FULL PIPELINE
# ---------------------------------------------------------
def process_drawing(image):
    if image is None:
        raise gr.Error("Please upload a drawing.")
    
    caption = generate_caption(image)
    tale = generate_fairy_tale(caption)
    audio = text_to_speech(tale)
    
    return caption, tale, audio

# ---------------------------------------------------------
# GRADIO INTERFACE
# ---------------------------------------------------------
with gr.Blocks(title="Fairy Tale from Child's Drawing") as app:
    gr.Markdown("""
    ## 🌈 Magic Storyteller for Kids  
    Upload a child's drawing → Get a short fairy tale → Listen to it!
    """)

    with gr.Row():
        img_input = gr.Image(type="pil", label="Child's Drawing")
        audio_output = gr.Audio(label="Narrated Fairy Tale")

    caption_output = gr.Textbox(label="AI Description of the Drawing")
    tale_output = gr.Textbox(label="Generated Fairy Tale", lines=4)

    generate_btn = gr.Button("✨ Create Story")

    generate_btn.click(
        fn=process_drawing,
        inputs=[img_input],
        outputs=[caption_output, tale_output, audio_output]
    )

# Запуск
if __name__ == "__main__":
    app.launch()