Alibaba's Wan 2.7 Image brings pixel-perfect hex colors and real human faces

Anyone who generates visuals with AI knows the drill: faces look too smooth and plasticky, brand colors drift into random shades, and text usually turns into unreadable alien glyphs.

Alibaba rolled out Wan 2.7 Image on April 1, bringing an all-in-one approach to image creation and precision editing that targets those exact design headaches.

Instead of treating prompt-based generation and photo editing as separate tools, Wan 2.7 Image consolidates both into a single model architecture. The goal is straightforward: build, tweak, and polish professional visuals without hopping between platforms.

Ditching the plastic look and locking exact hex palettes

Alibaba claims the new model takes a deliberate turn toward lifelike human faces with natural variation and real-world imperfections.

Earlier models earned a reputation for spitting out overly symmetrical, uncanny faces. Wan 2.7 Image lets users steer facial structure down to the details: bone structure, eye shape, contours, makeup, hair, ethnicity, age, and body type can all be fine-tuned through prompts.

Color control gets a massive upgrade as well. Creators can feed up to eight exact hexadecimal color codes directly into their prompts along with target proportions. For brand designers who need to protect strict corporate identity palettes, this eliminates the guesswork of regenerating until the blue looks right.

Long text rendering and 12-shot batch consistency

Text rendering inside images has also taken a leap forward. Supported by a long-context framework, the model can render up to 3,000 tokens of readable text across more than 12 languages.

Alibaba showcased examples resembling high-res academic paper pages, complete with crisp typography, formulas, and structured tables. On top of that, an intuitive click-to-edit feature lets users highlight a region to move, align, or add elements with pixel-level precision.

For creators working on storyboards, comics, or product catalogs, the system can digest up to nine reference images and generate up to 12 stylistically cohesive shots from a single prompt.

Alibaba also unveiled Wan 2.7 Image-Pro, tuned for sharper prompt adherence, rock-solid composition, and native 4K output. Both tiers are available via Alibaba Cloud's Model Studio and the official Wan site, with integration coming to the Qwen flagship app.

Why it matters

The AI image generation space is shifting away from generic novelty art and moving straight into enterprise production pipelines.

Features like hex code locking and 3,000-token text rendering show Alibaba is paying attention to commercial friction points. In real-world ad agencies and e-commerce studios, having an AI produce a pretty picture is useless if the brand colors are off by two shades or the headline can't be rendered cleanly on the canvas.

We think this is a smart, utilitarian play. Pairing multi-image generation with reference-driven consistency connects neatly with Alibaba's broader video ecosystem, hinting at an end-to-end storyboard-to-video workflow.

Whether the long-form text rendering stays flawless across edge-case typography remains to be tested under heavy studio loads. Still, for commercial designers tired of constant manual cleanups, Wan 2.7 Image looks like one of the most practical additions to the creative stack this year.