Wan

Wan (万相) is Alibaba Cloud’s flagship generative AI platform for images and video, anchored by the Wan 2.7 model suite. Positioned at the intersection of creative production and enterprise AI infrastructure, it offers text-to-image, image-to-image, text-to-video, and video-to-video capabilities alongside advanced editing tools. The platform targets professional creators, marketing agencies, e-commerce operators, and developers who need fine-grained control over visual outputs rather than simple one-shot generation. Unusually for a major cloud vendor, Alibaba Cloud has open-sourced the Wan models, hosting weights on GitHub while simultaneously offering a managed API and web interface through its Alibaba Cloud console. This dual delivery model means teams can either self-host for maximum data sovereignty and customization or consume the service via API for scalability. The marketing emphasis is on controllability—precise color proportions, multi-image reference fusion, pixel-level editing, and temporal continuity in video—suggesting a product philosophy that treats AI as a production tool rather than a novelty. With support for twelve languages, chart rendering, and complex video narrative extension, Wan is clearly designed for global commercial use cases ranging from advertising campaign development to interactive media production. Its integration into the broader Alibaba Cloud ecosystem also implies access to enterprise-grade security, storage, and billing infrastructure, though this comes with the expectation that users will navigate Alibaba’s account and console framework. For organizations already embedded in Alibaba Cloud services, Wan presents a natural extension of their existing stack, while the open-source release lowers the barrier for academic researchers and independent developers who want to experiment with or fine-tune the underlying architecture.

Users access Wan through two primary channels: a browser-based creative console labeled “立即体验” (Try Now) and a developer-facing API platform. In the web UI, creators select a modality—such as image generation, video editing, or portrait customization—and provide prompts alongside optional reference assets. For image tasks, users can upload up to nine reference images to guide composition, style, and subject consistency, or use the interactive box-selection tool to restrict edits to specific regions. For video, the system accepts multi-dimensional instruction prompts that modify plot, environment, and camera work, while the temporal extension feature lets users anchor generation to a starting or ending frame—or both—to ensure narrative continuity. The “creative video replication” mode analyzes a reference video’s motion dynamics, effects, and camera movements and applies them to new content. Developers consuming the API integrate these same capabilities programmatically, likely through standard REST endpoints authenticated via Alibaba Cloud credentials, though the scraped content did not specify protocol details. Outputs appear to be managed within the Alibaba Cloud ecosystem, implying that storage, billing, and quota management are handled through the vendor’s existing cloud console. The workflow is inherently asset-heavy: instead of relying solely on text prompts, Wan expects users to supply reference images, frame anchors, or source videos to steer the model. This design rewards users who maintain organized media libraries and understand visual storytelling structure, but it may feel complex to novices accustomed to single-prompt generation. Once assets are uploaded, the model processes them through the selected pipeline—image fusion, video continuation, or portrait customization—and returns results that can be iteratively refined using the same control tools.