TikTok Owner’s New AI Tool Makes Lifelike Videos From A Single Photo

ByteDance has introduced OmniHuman-1, an AI system capable of generating lifelike videos of people speaking, gesturing, and performing various actions from just a single photo. The tool significantly outperforms existing methods in creating highly realistic human videos based on minimal input, especially audio.

Researchers showcased OmniHuman-1’s capabilities through sample videos, including a realistic animation of Albert Einstein speaking. Experts like USC professor Freddy Tran Nager praise its potential for entertainment and education but also acknowledge ethical concerns. The tool strengthens ByteDance’s position in the AI-generated video race, competing with digital influencers, virtual assistants, and AI-generated political figures.

NYU professor Samantha G. Wolfe warns of potential risks, such as AI-generated misinformation, as these videos become more realistic and influential. ByteDance trained OmniHuman on over 18,700 hours of human video data, though specifics remain unclear. Critics highlight privacy concerns, as TikTok users’ content might be used to improve AI-generated avatars.

Link to OmniHuman-1

Hugging Face launches AI App Store with over 400k Apps

Just took a look at Hugging Face Spaces, and wow. It's like stepping into a secret lab where the coolest AI experiments are happening! 🧪🤖

What is it?

A super simple way to create, host, and share AI apps – no complex setup required. Think of it as your personal AI playground! 🏗️

Why should you care?

✅ Build and showcase your AI apps effortlessly
✅ Share your creations with the community & get feedback
✅ Explore ~400,000 AI apps built by others – no coding required!

Who can use it?

👨‍💻 Developers – Deploy & host AI apps in minutes
🧑‍🚀 AI Enthusiasts – Try out AI apps with zero setup

The best part? No downloads needed – just fire up your browser and start exploring! 🔥

The AI App Store is here – whether you're building, discovering, or experimenting, Hugging Face Spaces is the place to be.

Link to Hugging Face Spaces

Deep Dive into LLMs like ChatGPT by Andrej Karpathy

Andrej Karpathy just released a 3h31m deep dive into the entire training pipeline of large language models. Whether you're a developer, researcher, or just AI-curious, this course breaks down:

🔹 Pretraining – Data, tokenization, Transformer internals, and inference with real-world examples (Llama 3.1)
🔹 Supervised Fine-Tuning – Why LLMs hallucinate, how they use tools, and quirks like memory and spelling
🔹 Reinforcement Learning – RLHF, DeepSeek-R1, AlphaGo, and how iterative improvements refine models

Karpathy makes complex AI topics digestible without oversimplifying. If you’ve ever wondered how these models "think" and why they behave the way they do this course is a must-watch.

Check it out: Deep Dive into LLMs

Free Alternative to OpenAI Deep Research - Meet OpenDeepResearcher

A new open-source AI system is here to automate comprehensive research reports. 🔍📄

Here’s how OpenDeepResearcher works:
✅ Performs searches based on your query
✅ Extracts key insights from results
✅ Initiates deeper searches dynamically
✅ Iterates until it gathers enough data
✅ Synthesizes everything into a detailed report

💡 The best part? Parallel processing makes it lightning-fast!

While it may not yet match OpenAI's deep research tool, it’s already a solid step toward future agentic research systems.

Check it out here: 👉 GitHub Repo

Building Apps on your Phone with Replit Agent for Free

Replit Agent is an AI-powered tool that allows users to build applications directly from their phones, making software development more accessible. By simply inputting an idea, the AI generates a prototype within minutes.

Replit aims to revolutionize coding on the go, positioning the Replit Agent as a key step toward human-AI collaboration. Currently in early access, it is available to Replit Core and Teams subscribers at no extra cost, with pricing updates expected later in 2024.

To use Replit Agent, users provide a detailed prompt, allow the AI to select technologies, review the generated plan, and collaborate with the agent to refine and deploy their application.

Link to documentation