Wan 2.7 Pro Edit

• 🆕 Alibaba's Pro-tier image editing model in the Wan 2.7 line
• 🎯 Instruction-based editing plus masked inpainting and outpainting
• 👁️ Accepts multiple reference images for guided composition
• 📚 Renders text and integrates content across combined inputs
• 🏢 Built and served through Alibaba Cloud Model Studio
• 🔧 Tuned for high editing precision over the base Wan 2.7 editor
• ⚡ Released April 2026 as part of the wan2.7-image family
• 🌐 Handles both Chinese and English editing instructions

Alibaba

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Wan 2.7 Pro Edit is the editing-focused, Pro-tier member of Alibaba's Wan 2.7 image stack, designed for inpainting and instruction-driven image manipulation rather than generation from scratch. It corresponds to Alibaba Cloud's wan2.7-image-pro editing endpoint, which the documentation positions for scenarios requiring high editing precision or generation of multiple coherent images. Core capabilities include instruction-based editing (modifying content without specifying a region), masked inpainting, multi-reference composition, and combining elements across input images.

Within the family, it pairs naturally with Wan 2.7 Pro for generation and sits alongside Alibaba's parallel Qwen editors such as Qwen Image 2 Pro and Qwen Edit 2511. The "Pro" designation reflects a higher-capability, higher-cost tier than the standard wan2.7-image editor, per the API references.

Because this is the first edition in the wan-pro-edit family, there is no earlier same-family release to benchmark against; it inherits the Wan 2.7 generation's instruction-following editing approach rather than carrying forward a prior Pro Edit version. Alibaba publishes the model through Model Studio with options for resolution up to 2K and watermark toggles, supporting public-URL, local-file, and Base64 image inputs. No official or independent benchmark figures for this specific editing variant were available at the time of writing, so capabilities here are described from primary documentation only.