The open-source artificial intelligence video generation landscape has seen a significant advancement with the recent release of Wan 2.2, developed by Tongyi Lab, an Alibaba Cloud initiative. This new model is demonstrating performance levels comparable to, and in some cases surpassing, established closed-source models like Kuaishou's Kling 2.1, MiniMax's Hailuo, and Google's Veo. The development marks a pivotal moment for the accessibility and capability of open-source AI tools.
AI researcher Ashutosh Shrivastava recently highlighted this development on social media, stating, > "Wan 2.2 just dropped, and I tested it against the best video models: Kling 2.1, Hailuo 02, and Veo3. The open-source model is now seriously on par with closed-source ones." Shrivastava's assessment included a test prompt featuring an "Anthropomorphic white cat walking down a fashion runway," emphasizing the model's ability to handle complex and creative scenarios.
Wan 2.2, officially released on July 28, 2025, leverages a Mixture-of-Experts (MoE) architecture, which contributes to its enhanced efficiency and ability to produce cinematic-quality videos. This model is notable for its compatibility with consumer-grade GPUs, making high-quality video generation more accessible to a broader range of users and developers. It supports both text-to-video (T2V) and image-to-video (I2V) functionalities.
In contrast, Kling 2.1, a closed-source model from Chinese tech giant Kuaishou, was released around May-June 2025 and offers various quality tiers up to 1080p. It is recognized for its realistic motion, physics simulation, and strong adherence to prompts, primarily excelling in image-to-video generation. Hailuo, developed by Chinese AI company MiniMax, is another closed-source contender known for generating high-quality, realistic videos, particularly with human-like characters, from text prompts.
Google's Veo, also a closed-source offering, is touted as the company's most advanced generative AI model for creating 1080p videos from diverse prompts, incorporating sophisticated cinematic controls. Historically, closed-source models have often led in performance due to dedicated development teams and proprietary data. However, the rapid evolution of open-source solutions like Wan 2.2 is challenging this dynamic, fostering greater transparency, flexibility, and cost-effectiveness within the AI community.
The increasing performance parity between open-source and closed-source AI video models signals a maturing industry where innovation is becoming more democratized. This trend is expected to accelerate further advancements and applications across various sectors, from independent content creation to professional media production.