On July 28, 2025, Wan 2.2 Video has been released and was capable of state-of-the-art video generation. On July 29 2025, we optimized it to make it the fastest video generation model, and deployed it on Replicate to make it accessible for everyone! You can now use the final Replicate text-to-video endpoint and image-to-video endpoint to generate videos in seconds!

Juicing Wan 2.2!

How does Wan 2.2 Juiced work?

Wan 2.2 Juiced endpoints are built upon the Wan 2.2 model, which introduces a novel dual-expert architecture. It combines a high-noise expert for the early stages, which focuses on overall layout, and a low-noise expert for the later stages, refining video details. For a deeper dive, you can explore the full tech report from the Wan team here.

How is Wan 2.2 Juiced optimized and deployed?

To optimize performance, we applied the most advanced compression methods from the Pruna package before deploying Wan 2.2 Juiced via Replicate, enhancing both speed and efficiency.

What is Wan 2.2 Juiced good for?

Wan 2.2 Juiced is designed for high-quality video generation with precise control over cinematic aesthetics. It incorporates specially curated data with fine-grained labels for lighting, composition, and color, enabling customizable video styles to meet a wide range of creative preferences. See for Yourself below!

The Fastest Video Generation Model!

We have optimized Wan 2.2 Juiced for speed. After compression, Wan 2.2 Juiced generates 480p and 720p videos from one image in 28s and 80s, and generates a 480p video from text in 33s. This represents an acceleration of 1.5x-2x compared to the base models. Note, this acceleration can be enhanced even further by distributing on more hardware, especially for the wan-2.2-i2v-480p-fast and the wan-2.2-t2v-480p-fast endpoints, which only use a single H100.

Beyond time savings, we also released this model with Replicate at competitive pricing. For the image-to-video model, the pricing is $0.05 for a 5s video in 480p and $0.11 for a 5s video in 720p. For the text-to-video model, the pricing is $0.05 for a 5s video in 480p. Beyond being the most affordable Wan 2.2 endpoints on the market, it makes video generation more accessible for everyone. To put it into perspective, the text-to-video Wan2.2 Juiced endpoint is 100x cheaper than Veo3, at $0.06 instead of $6.00 per video!

Wan 2.2 has amazing quality, as shown on Wan tech report, outperforming Seedance, Hailuo, Kling, Sora. We went a step further and created the “Frame Arena” tool to compare two generated videos frame by frame. It enables comparing videos generated by Wan 2.2 and Wan 2.2 Juiced. While you can upload your own video, we pre-uploaded some videos showing that Wan 2.2 Juiced achieves state-of-the-art quality comparable with Wan 2.2 while being significantly faster for video generation.

Quality evaluation of Wan 2.2 against competitors from Wan tech report. To get something more interactive, make sure to look at a comparison of the generated videos by using the “Frame Arena” tool.

Disclaimers

While we are excited to release this practical research implementation, we want to emphasize a few essential disclaimers regarding the techniques used to enable this release:

• Original Work: The Wan 2.2 endpoints are built upon the original work of the Wan-Video team. The base model's limitations and licensing terms remain applicable to this adaptation.

• Test in Your Use Case: We recommend thoroughly testing the model in your specific use case, validating your outputs, and reporting any unexpected behaviors or limitations to us.

In short, have fun with the model, make the most of its capabilities, and help us improve it when needed. Share your findings and use cases on social media, and we’d be happy to reshare!

Enjoy the Quality and Efficiency!

Want to take it further?

• Try our Replicate text-to-video endpoint and image-to-video endpoint with just one click.

• Compress your own models with the Pruna package.

• Stay up to date with the latest AI efficiency research on our blog, explore our materials collection, or dive into our courses.

Give us a star to show your support!