Alibaba Unveils Wan2.7-Image to End the Era of Generic AI Faces

Ali's official large model team has launched the unified image generation and editing model Wan2.7-Image today. This model represents a qualitative leap forward from its predecessor in portrait customization, color control, and long-text rendering, aiming to overcome the generic "sameness" often seen in AI-generated visuals.

The model is now available for access via API on the Alibaba Cloud BaiLian platform and is also open for public trial on the WanXiang website.

Core Enhancements: Virtual "Facial Sculpting" and Precision Color Control

Wan2.7-Image introduces several industry-leading, differentiated features that significantly expand creative possibilities:

Unique Faces for Every Subject: It features enhanced virtual "facial sculpting" capabilities. Users can precisely control bone structure, eye shape (like almond or phoenix eyes), and other facial details through text prompts, effectively eliminating standardized AI-generated faces.

Precision Color Palette: It newly incorporates advanced color control. Users can extract color ratios from reference images to accurately replicate specific hues—such as Van Gogh's vibrant yellow or Picasso's cool blue—into new creations.

3K Token Ultra-Long Text Rendering: It addresses common AI text-generation issues, supporting up to 12 languages. The model can render complex text, tables, or formulas spanning an entire A4 page with print-ready quality.

In the realm of image editing, Wan2.7-Image debuts an "interactive editing" function. Users can simply select a precise area to add, align, move, or even perform pixel-level logical replacements—like swapping ice cubes for fruits while perfectly preserving the scene's original lighting.

Furthermore, the model supports generating coherent sets of up to 12 images, maintaining high consistency in style and subject characteristics across different scenes—ideal for group portraits or coordinated product combinations.

Ali explains that Wan2.7-Image employs a unified architecture for both generation and understanding. By achieving semantic mapping within a shared latent space, the model moves beyond guessing pixel patterns and develops a genuine underlying comprehension of semantics.

With its release poised for the 2026 visual creation landscape, Wan2.7-Image signals a shift in AI image generation: from "random card drawing" to "industrial-grade" precision control. This high-fidelity editing capability is set to dramatically lower the barrier for professional content creation, whether for film storyboards, e-commerce advertising, or social media transformations.