---
title: MakeMeDinner AI
emoji: 🍳
colorFrom: red
colorTo: yellow
sdk: streamlit
sdk_version: 1.40.0
app_file: app.py
pinned: false
license: mit
---

# MakeMeDinner AI — Multimodal Cooking Assistant

**AMD Developer Hackathon · Vision & Multimodal AI Track**

## The Full AI Pipeline

| Modality | AI Model | What It Does |
|----------|----------|--------------|
| **Vision In** | `vit-gpt2-image-captioning` (via HF Inference API) | Scans your fridge/pantry photo and identifies available ingredients |
| **Language** | `zephyr-7b-beta` (via HF Inference API) | Generates custom recipes from detected ingredients + your craving description |
| **Audio Out** | Browser `SpeechSynthesisUtterance` | Reads recipes aloud step-by-step as you cook |
| **Vision Out** | `FLUX.1-schnell` / `SDXL` (via HF Inference API) | Generates photorealistic food images for each recipe |

## How It Works

1. **📷 Scan** — Upload a photo of your fridge or pantry. Vision AI identifies eggs, tomatoes, spinach, cheese, etc.
2. **🧠 Generate** — Type what you're craving ("spicy Mexican", "cozy Italian"). Zephyr-7B crafts a custom recipe.
3. **🔊 Cook** — Tap "Read Aloud" and the browser TTS walks you through every step hands-free.
4. **🎨 Visualize** — Generate a photorealistic food photo of your dish with FLUX.1 to share or inspire.

## Tech Stack

- **Streamlit** — UI framework
- **HuggingFace Inference API** — Vision, LLM, and text-to-image models (zero local GPU)
- **Pillow** — Image processing
- **Browser Web Speech API** — Text-to-speech

## Setup

```bash
pip install -r requirements.txt
HF_TOKEN=your_token streamlit run app.py
```

Requires a HuggingFace token with Inference API access for full multimodal features.