Google's Veo 3: How the New Video Generator Surpassed Sora and Kling

In early January 2025, Google introduced Veo 3, a new generation of video generator that topped the independent Artificial Analysis rankings within just a few days. The model outperformed OpenAI's Sora Turbo, Kling 1.6, and Hailuo Video 2.0 in terms of physics, motion realism, and scene detail. While Veo 3 is currently only available in experimental mode via Google Labs, it is already clear that competition in the AI video market has reached a new level.
What Veo 3 Can Do
The main difference between Veo 3 and its competitors is its improved object physics and natural character movement. While Kling 1.6 sometimes showed artifacts during rapid camera angle changes and Hailuo 2.0 struggled with facial expressions, Veo 3 handles both tasks significantly better. The model generates clips of up to 10 seconds in 1080p resolution, supports both vertical and horizontal formats, can work with image references, and understands complex prompts involving multiple actions within a single frame.
Another key feature is camera control. You can specify the type of movement in your prompt: pan, zoom, orbit, or static shot. Previously, this was only seen in Runway Gen-3 Alpha, which required a $95-per-month subscription. In Veo 3, this capability is built-in from the start, although the model is currently in closed testing and not available to everyone.
Veo 3 vs. Competitors: Where the Difference Lies
To understand how the new model differs from Kling, Hailuo, and Sora, it is enough to look at three key scenarios:
- Dynamic scenes with people. Veo 3 better maintains body proportions during movement and does not confuse limbs when a character turns their back to the camera. In such shots, Kling 1.6 sometimes generates extra fingers or blurred outlines.
- Complex lighting. When shooting at twilight or with multiple light sources, Veo 3 correctly renders shadows and highlights. In similar conditions, Hailuo 2.0 can overexpose the background or make shadows too harsh.
- Text in the frame. While no model can generate readable text perfectly yet, Veo 3 handles it better than the rest: letters do not drift and remain in place even when the camera moves.
In terms of generation speed, Veo 3 is slower than Hailuo—a clip takes about 40 seconds to create compared to 25 seconds for its Chinese competitor. However, the quality of the final result is higher, especially when precise detail or realistic physics for liquids and fabrics are required.
Who Is Veo 3 For and What Is It Used For?
The new model addresses several practical tasks that previously required either expensive subscriptions or a combination of multiple tools:
- Video advertising prototyping. Marketers can assemble a draft video with the desired mood and angles in a couple of minutes to show a client a concept before filming.
- Social media content. Short vertical videos for Reels, Shorts, and TikTok are generated from a single prompt—no need for editing or searching through stock footage.
- Educational materials. If you need to show a process or a product demonstration, Veo 3 creates an illustrative video without the need for a camera operator or editor.
- Creative experiments. Designers and video artists use the model to explore visual solutions—generating a dozen scene variations is easier than drawing a storyboard by hand.
For now, Veo 3 is not suitable for long videos—the 10-second limit does not allow for filming full-length footage for YouTube. But for quick tasks where image speed and quality are paramount, the model is already ahead of most alternatives.
Try Video Generation at GEMERA AI
Want to test the best video generation models without VPNs or complex setups? GEMERA AI offers access to Kling, Hailuo, and other current tools—start generating directly from our Telegram bot or web dashboard. Compare the results of different models on your own tasks and choose the one that works best for you.