Alibaba's generative models come out of Tongyi Lab, the AI research division of Alibaba Cloud. The lab develops the Qwen family, the brand Alibaba has unified its large-model work under, alongside the Wan video models, originally launched as Tongyi Wanxiang. Many are released open-weight under permissive licenses, making them a popular foundation for developers and the open-source community.
On the visual side, Qwen-Image is a text-to-image foundation model known for state-of-the-art complex text rendering, while Qwen-Image-Edit brings precise, instruction-driven image editing.
In video, the Wan series spans text-to-video and image-to-video generation, with Wan 2.5 introducing native synchronized audio and later versions adding reference-to-video, multi-shot control, and flexible aspect ratios.
Notable model families on fal:
- Happy Horse: text- and image-to-video, with native audio generation
- Qwen-Image: text-to-image with leading text rendering
- Qwen-Image-Edit: instruction-based image editing
- Wan 2.1 / 2.2 / 2.5 / 2.6: text- and image-to-video, some with native audio
- Wan-Animate: character animation driven by reference video
