Video2Recipe

Video2Recipe extracts structured recipes from cooking videos using multimodal analysis of speech, on-screen text, and visual cues. Built with .NET and C# to turn unstructured video content into clean, editable recipe data.

Overview

Cooking videos often spread recipe information across multiple channels: spoken instructions, on-screen text overlays, captions, and visible ingredients or actions.
Video2Recipe combines these signals to generate a structured recipe draft that can be reviewed, edited, and saved.

Goals

• Extract recipes from general video content, not just a single platform.
• Combine audio, OCR, and visual analysis into one recipe extraction pipeline.
• Produce structured, editable recipe data instead of raw summaries.
• Keep the architecture modular so providers and extraction components can be swapped easily.

Planned Features

• Video upload and processing
• Audio extraction and transcription
• OCR for on-screen ingredient and step detection
• Visual cue analysis from selected video frames
• Structured recipe generation
• Confidence scoring and evidence tracking
• Manual review and correction UI
• Recipe export for downstream usage

Tech Stack

• .NET
• C#
• ASP.NET Core
• Blazor
• FFmpeg
• PostgreSQL
• AI/LLM integration for extraction and normalization

Project structure

Video2Recipe.sln
|-- CrossCutting
|   |-- DataClasses
|-- Data
|-- Infrastructure
|   |-- DiMappings
|-- Logic
|   |-- Business
|   |-- Domain
|-- UI
|   |-- WebClient

Current Status

This project is in an early planning/MVP phase.
The first milestone is a local end-to-end flow for uploaded cooking videos: video upload -> audio extraction -> transcription -> recipe draft generation.