What Is Hugging Face and How To Use It

Overview

This video introduces Hugging Face as an AI platform and walks through how to use it to:

• Try existing models
• Build and customize your own AI apps
• Deploy and/or run them locally

---

What Hugging Face Is (Platform Overview)

Hugging Face is described as an open-source AI community and a global hub for open-source large language models, including models from companies such as:

• Meta
• Google
• DeepSeek
• Alibaba

It’s positioned as a place where companies publish new AI models first. The video also claims there are 1.5M+ open-source AI models available.

Cost notes

• It’s free to try (for many features)
• GPU usage can require payment, with the mentioned plan price being $9/month (the speaker subscribes)

---

Main Hugging Face Sections (Core Product Features)

1) Spaces (Best for Non-Technical Users)

Spaces is a page of ready-to-run apps categorized by type, such as:

• Image generation
• Video generation
• Text/language generation
• Speech synthesis
• 3D modeling

It includes discovery features like:

• “Spaces of the Week”
• “Trending running apps”
• Filtering by space type

Demo examples

• Relighting

  • Upload an image
  • Choose a background
  • Click Relight
  • Get a before/after result where the object is blended into the new background in ~3 seconds (intended to look more realistic)

• Sesame CSM

  • Generates/uses two speakers
  • Speaks user-provided text with a voice persona-like effect
  • The speaker notes it’s not perfect, but it’s quickly testable

---

2) Model Hub

The Model Hub provides access to models categorized by domain, such as:

• Computer vision
• NLP
• Audio

Models are shown as open source and can be sorted—trending is the default view.

Model pages also link to:

• Spaces that use the model
• Options to deploy or use the model
• A community/discussion area

---

3) Datasets

Datasets are described as inputs for model training and fine-tuning.

Key capabilities mentioned:

• Filter by language and task
• Inspect dataset contents
• Run SQL queries (as mentioned)
• Download datasets

The emphasized use case is fine-tuning your own model.

---

Tutorial: Building and Customizing an AI Image Editing App

The speaker demonstrates building an app using the latest Google Gemini update via the Gemini API.

Practice demo

• Upload an image (e.g., a girl)
• Provide an editing prompt (e.g., make her eat french fries instead of holding something)
• The demo mentions trying the app for free initially
• Later, you can use your own Gemini API key if desired

---

How to Customize and Run Your Own Version

The workflow described:

1. Duplicate the Space / repo
2. Add your secret (API key) and duplicate
3. For self-hosting/customization, use the public code
   • Download/clone the repo
   • Clone into Cursor (the code editor mentioned)
4. Use Cursor Agent to help step-by-step:
   • Create a Python virtual environment
   • Install requirements
   • Configure the Gemini API key
5. Run the app locally:
   • It serves a local URL
   • After providing an API key, prompting works and image generation/editing begins

---

Example Customization Prompt

The speaker shows changing the output aesthetic/style through prompt instructions, for example:

• “Sailor Moon colors and cute fonts”

After accepting changes, the app is run again to produce different outputs.

Note: The speaker mentions considering licensing/credits when reusing shared code.

---

Overall Takeaway / Intent

Hugging Face is presented as:

• A large library (Model Hub)
• A discovery/testing platform (Spaces)
• A training resource (Datasets)

It also provides a practical way to prototype and host custom AI apps using shared repos and API keys.

---

Main Speakers / Sources

• Speaker: “Hey guys” host (single speaker; no other named co-hosts)
• Primary referenced product/system: Hugging Face (Spaces, Model Hub, Datasets)
• External model/API referenced: Google Gemini (via Gemini API key)