April 1 News.AlibabaImage GenerationThis post is part of our special coverage Global Voices 2011. In response to the pains of aesthetic fatigue and colour out of control in the current AI biography, Wan2.7-Image brings with it a more “living sense” of character generation, precision colour control and super-long text rendering。
According to the presentation, the model could"GOODBYE AI STANDARD FACE" AND "THOUSAND ONES", support brand-new palettes, precision control colours; support 3K token super long text rendering, A PAGE FULL OF A4 PAPER。
In addition, the model brings with it an interactive editing of “where there is no good, where there is no good,” and has the capability of a full link, such as a texture, a map, an image command editing and interactive editing。
Wan2.7-Image enhances virtual face squeezing and supports all-round customizations from bone, eyes to five nuances, such as changing face shapes in hints (e.g. goose and egg face, round face, square face, long face, etc.), eye features (almond eye, deep eye nest, round eye, Dan Fong eye, etc.) and achieving “thousands of faces”。
Wan2.7-Image is completely new to support the " palette " function, with a user extracting or entering in one key the various colours and percentages of the reference diagram, and is free to regulate the number and percentage of colours and customize the colour scheme。
Models can be rendered as long as they are printed, supporting 12 languages, with a maximum of 3K token input of extremely long text, and can output a paper on A4 paper。
1AI understands that it also supports interactive editing of “where there is no good” . Aligns pixel-level intent through precision frames, adding, aligning, moving elements or logo in the specified area。
It also supports multi-subject coherence: a maximum of 9 graphics, film posters and furniture combinations to maintain consistency between style and features。