On April 1, 2026, Alibaba's Tongyi Qianwen team officially released the unified image‑generation and editing model Wan2.7‑Image. This model directly tackles the toughest problems in today's AI image field — aesthetic fatigue from "standard faces", runaway color, and unstable instruction following — achieving truly "thousand‑faces" portrait customization and marking a shift in AI visual generation from "templated" to "personalized". Users can experience it for free or via API at Tongyi's site (tongyi.aliyun.com/wan), wan.video and the Alibaba Cloud Bailian platform.
Wan2.7‑Image is not a simple iteration but an integrated image‑generation and editing foundation model built on Alibaba's Wan series. It supports text‑to‑image, image‑to‑image sets, image‑instruction editing and interactive editing across the full pipeline, making breakthroughs especially in three dimensions: portrait generation, color control and long‑text rendering. Official blind tests show it ranks first domestically in human preference scores; the generated images convey a stronger sense of "real people", decisively leaving behind the previous AI portrait problems of homogenization, plastic look and limited angles.
Core innovation: from "one face fits all" to "thousand faces"
- Thousand‑faces face‑customization system: This is Wan2.7‑Image's most striking upgrade. The model enhances virtual character customization, supporting precise control of bone structure, eyes, facial features and other details via prompts. For example, users can specify traits like "oval face", "almond‑shaped eyes", "phoenix eyes", "deep eye sockets", and even fine‑tune facial proportions to create highly personalized characters. Compared with previous versions, Wan2.7‑Image shows significant improvements in portrait consistency, multi‑angle generation and naturalness of expressions; generated people appear more lifelike and culturally diverse, breaking the common "influencer face" template of AI.
- Precise color‑palette function: To address the "color blind box" problem, the model introduces a "palette" mechanism that accepts exact color codes and ratio controls, letting users manage overall tones, local highlights and ambient light as if mixing a palette. Whether it's brand VI colors, cinematic lighting or a specific artistic style, the model can reproduce them reliably, greatly reducing post‑editing costs.
- Long‑text rendering and interactive editing: The model supports rendering complex prompts of over 4,000 English characters (or equivalent Chinese), with marked improvements in text clarity and layout stability. It also provides precise box‑selection editing and interactive modification capabilities, allowing users to adjust local areas and produce consistent multi‑image outputs without re‑running full generations. The model accepts up to nine reference images, enabling album outputs with high consistency in style and subject.
These innovations stem from Alibaba's optimizations at the infrastructure level, including more refined diffusion‑model training data and multimodal alignment techniques, enabling Wan2.7‑Image to maintain high generation quality while significantly improving instruction adherence and controllability.
Availability and access
Wan2.7‑Image is fully deployed on the Alibaba Cloud Bailian platform and available via API; it is also integrated into the Tongyi Qianwen App and Quark's "Zaodian" AI creation platform for zero‑barrier user access. The model supports multiple resolutions and styles and is suitable for scenarios such as short‑drama storyboarding, e‑commerce main images, social‑media covers, children's picture books and academic illustrations. Some features are available for free trial, while enterprise users can achieve batch production through the API.
Far‑reaching impact: democratizing visual creation and reshaping industry
For content creators: Wan2.7‑Image significantly lowers professional barriers. Independent media and short‑drama teams can quickly generate personalized characters and multi‑style covers; e‑commerce practitioners can produce diverse scenes, close‑ups and product highlight images from a single model photo, saving substantial shooting costs; educators and researchers can easily create infographics and illustrations. The aesthetic fatigue caused by "AI faces" is effectively resolved, allowing creators to focus on creativity rather than repeatedly tuning prompts.
For the industry ecosystem: In short dramas, film, and advertising, Wan2.7‑Image's face‑customization and storyboard capabilities enable low‑cost previews and VFX testing, accelerating the "AI+film" integration. The model also supports Skill calls, expanding into game NPCs, virtual anchors and other scenarios. Overall, it accelerates the transformation of AI image generation from a "toy" to a productivity tool.
For global AI competition: As a domestically leading image‑generation model, Wan2.7‑Image offers differentiated advantages in portrait diversity and controllability, complementing international competitors like Midjourney and DALL·E. China's "small‑to‑win‑big" approach in visual AI is again validated, providing more accessible options for developers worldwide.
Challenges and outlook: Despite leading performance, Wan2.7‑Image as a new model still needs ongoing optimization for stability in extremely complex scenarios. In the future, Alibaba may open more open‑source releases or expand video linkage capabilities to promote a full‑chain visual creation ecosystem.
The release of Wan2.7‑Image is not only a targeted push by Alibaba in the AI visual track but also a positive response to industry demand for "thousand‑faces" personalization. It makes AI imagery no longer a "standard template" but a creative partner that understands users and creativity. As generative AI becomes more widespread, this model is opening a new era of visual content production — where everyone can easily have a unique "AI face".
▶️ Enjoy this article? Creating takes real effort!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares are our greatest support! 🙏
More AI news at: https://www.airosetta.com