Whoa. Just when you think you’ve caught up, the world of AI video makes a giant leap. It’s moving so fast, it’s hard to keep track!
Luckily, I just saw an incredible post from Tianyu Xu that gives a perfect snapshot of where we are right now. He’s been tracking this for two years, and his summary is a must-read.
Here’s the breakdown of his findings that blew me away:
📌 The Core Tech
Tianyu explains that most tools fall into three camps:
- Text-to-video (the most common)
- Image-to-video
- Video-to-video
✅ What’s Possible TODAY
According to his analysis, the top models can already do some wild stuff:
- Generate consistent characters across clips.
- Produce clips up to 10 seconds long.
- Output in 1080p, with options to upscale to 4K!
⚙️ The Heavy Hitters
Tianyu shared a list of the current state-of-the-art models. This is the bleeding edge:
- Runway Gen-4
- Luma Ray2
- DeepMind Veo 2
- Kling v2.0
- OpenAI Sora (of course!)
✨ The Smart Aggregators
He also points out another huge category: aggregators. These are tools that use multiple models to optimize for things like storytelling or creating ads. Super useful!
- OpenArt
- LTX Studio
- Freepik
🤖 And Don’t Forget AI Avatars!
Finally, he highlights the boom in realistic AI presenters from tools like HeyGen and Synthesia. You can create a talking head without ever stepping in front of a camera.
Tianyu calls this “creativity-democratization on steroids,” and I couldn’t agree more. The progress is just staggering.
He posted a video showing the difference between 2023 and 2025 AI video quality, and it’s mind-blowing. You HAVE to go see the original post from Tianyu Xu to get the full picture and watch his video!