README.md

metadata

title: Bilingual Storyteller Illustrator
emoji: 👀
colorFrom: gray
colorTo: purple
sdk: gradio
sdk_version: 5.18.0
app_file: app.py
pinned: false
license: apache-2.0

AI-Powered Bilingual Storyteller & Illustrator - Technical Summary

Core Functionality

• Generates stories in English and Arabic with emotional analysis and optional illustrations
• Uses template-based approach with AI models to ensure quality and safety

Technical Architecture

Story Generation

NLP Pipelines

• English Text Generation Pipeline:

  pipeline("text-generation", model="EleutherAI/gpt-neo-1.3B", device="cpu")
  

• Arabic Generation:

  # Uses MT5 instead of standard pipeline
  AutoTokenizer.from_pretrained("google/mt5-small")
  AutoModelForSeq2SeqLM.from_pretrained("google/mt5-small")
  

• Sentiment Analysis Pipelines:

  # English
  pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english", device="cpu")
  
  # Arabic
  pipeline("sentiment-analysis", model="CAMeL-Lab/bert-base-arabic-sentiment", device="cpu")
  

• Translation Pipelines:

  # Arabic to English
  pipeline("translation", model="Helsinki-NLP/opus-mt-ar-en", device="cpu")
  
  # English to Arabic
  pipeline("translation", model="Helsinki-NLP/opus-mt-en-ar", device="cpu")
  

Visual Generation

• Image Generation Pipeline:

  pipe = StableDiffusionPipeline.from_pretrained(
      "runwayml/stable-diffusion-v1-5",
      torch_dtype=torch.float16
  )
  

• Efficient GPU resource management via @spaces.GPU decorator
• Scene extraction from story content

Content Safety System

• Multi-layered content filtering
• Regex pattern detection for inappropriate content
• Repetition detection (unique word ratio < 0.4)
• Fallback mechanisms to reliable templates

Implementation Highlights

MultilingualStoryGenerator Class

• Central class managing generation in both languages
• Handles language detection, content safety, and sentiment analysis
• Template selection logic based on keyword matching
• Parameter extraction from prompts

Story Templates

• Three categories: Adventure, Friendship, Fantasy
• Multiple variations in both languages
• Dynamic parameter filling

GPU Resource Management

• @spaces.GPU decorator for efficient GPU allocation
• Pipeline moved to GPU only when needed for image generation
• Proper cleanup with torch.cuda.empty_cache() and gc.collect()

Error Handling

• Comprehensive logging system
• Graceful degradation for missing components
• Multiple fallback mechanisms

Technical Requirements

• Python 3.8+
• CUDA-capable GPU (for image generation)
• Key dependencies: transformers, diffusers, gradio, torch