Most image generation tools share a common trait that irritates professionals: the result has that «AI-generated» look. Generic faces, unpredictable colors, and text that turns into gibberish. Alibaba decided to tackle these problems systematically – and on April 1, 2026, it unveiled Wan2.7-Image, a unified model geared towards professional image work.
More Than Just «Generating a Picture»
Wan2.7-Image is positioned as an end-to-end tool: from creating an image based on a text description to editing and finalizing it. All of this happens within a single model, without the need to switch between different services.
Simply put, you can describe a scene with words, receive an image, adjust details directly within the interface, and export the result – all without leaving a single workspace. The model can also process multiple images simultaneously, which speeds up workflows, such as creating a series of materials for an ad campaign.
According to anonymized user tests, where participants selected the best result without knowing which model created it, Wan2.7-Image outperformed its competitors in visual quality, text rendering accuracy, and its understanding of complex visual concepts.
Characters That Don't Look Alike
One of the main complaints about AI generation is its «genericness.» All the faces look similar because the model averages its training data. Wan2.7-Image offers precise control over anatomical details: the shape of the cheekbones, the set of the eyes, and facial proportions. This allows for the creation of characters who look like specific, recognizable individuals, not like a «generic person from a stock database.»
This level of control is particularly important in illustration, game design, and advertising – any field where a character's visual identity is crucial.
Color by Code, Not «Approximately»
Another chronic problem for AI tools is color. Try to explain to a model that you need a very specific shade of blue – the one in your brand's style guide. This used to turn into a series of frustrating attempts.
Wan2.7-Image introduces a «color palette» feature: you can include specific color codes in the text prompt and specify the proportions in which to use each one. The model will reproduce them accurately, without interpretation. This means a company's brand colors or an artist's chosen palette will be rendered literally, not just approximately.
Text in Images – Finally Legible
Text within images has historically been a weak spot for all generative models. Letters bleed, words become distorted, and formulas turn into visual noise.
Wan2.7-Image approaches this task differently. The model supports text prompts up to 3,000 tokens long – that's about several pages of dense text. It can generate academic texts of typographic quality, complex formulas, and tables. Support for 12 languages is included.
This opens up possibilities that previously required manual work: infographics with readable captions, educational materials, and technical diagrams can now all be obtained directly from the model.
Editing by Selection, Not Guesswork
The editing interface deserves special mention. Wan2.7-Image implements a «click-to-edit» mode: the user highlights a specific area of the image and gives an instruction, such as to add an element, move an object, or change its position. The precision is down to the pixel level.
This is fundamentally different from how editing usually works in generative models, where altering one detail can often lead to unpredictable changes in other parts of the image. Here, the scope of the change is confined to the user's selection.
Scale and Storyboarding
For those working with batch content – like storyboards, architectural concepts, or catalogs – the model supports up to nine reference images at once and can generate up to 12 images in a single request. This allows for maintaining a consistent visual style across an entire series without manually configuring each image.
Pro Version with 4K and Stricter Prompt Following
Alongside the base model, Wan2.7-Image-Pro was also introduced. It offers more stable composition, more precise adherence to text prompts, and 4K resolution output. For those who require high final detail – for example, when preparing materials for print or for high-resolution displays – this is a substantial upgrade.
Context: The Wan Series and Its Evolution
The Wan series has existed since 2023 and has since gone through several iterations. Wan2.7-Image is the most feature-rich version to date. The models are available via Alibaba's cloud platform for developers and will also be integrated into the Qwen App, Alibaba's flagship AI application.
We will see how the model performs in real-world professional practice, but judging by its announced capabilities, Alibaba is seriously targeting the segment currently dominated by Western tools.
