Shulou(Shulou.com)11/24 Report--

Thanks to the CTOnews.com netizen Si Kong Zhe Yi, the best friend of the heart for the clue delivery! CTOnews.com October 26 news, according to the "goose factory technology school" news, Tencent mixed yuan model recently ushered in a new upgrade, and officially opened to the public "Wen Sheng Tu" function. Officials announced that the upgraded Tencent mixed Chinese language effect as a whole surpassed that of GPT3.5, and the code capacity was greatly improved by 20%, reaching the industry-leading level.

According to the official account of ▲ Tu Yuan Goose Factory Technical School, the difficulties of the large model Wensheng diagram are reflected in the semantic understanding of prompt words, the rationality of generating content and the effect of generating pictures. In view of these three technical difficulties, Tencent conducted special technical research and proposed a series of original algorithms:

In terms of semantic understanding, Tencent mixed element adopts a fine-grained model in both Chinese and English. The model is modeled in both Chinese and English to achieve bilingual understanding, rather than through translation. The optimization algorithm improves the model's perception and generation of details, and effectively avoids misunderstandings under multicultural differences.

In terms of content rationality, AI generates human structure and hands are often easily deformed. By enhancing the two-dimensional position perception of the image of the algorithm model, and introducing a priori information such as human skeleton and human hand structure into the generation process, the mixed-element textual graph makes the structure of the generated image more reasonable and reduces the error rate.

In the aspect of picture texture, mixed meta-textures are based on the method of multi-model fusion to improve the texture. After the optimization of the model algorithm, the effect of the mixed portrait model, including hair, wrinkles and other details, has been improved by 30%, and the effect of the scene model, including vegetation, ripples and other details has been improved by 25%.

▲ Touyuan Goose Factory Technical official account if users want to generate a more realistic picture, or close to a realistic picture, they can add a description such as "realistic" or "photographic style". If users want a specific style of picture, it is recommended to include that style description in the prompt, such as oil painting style, cyber punk style and so on. In addition, the user can describe the picture in as much detail as possible to achieve satisfactory results.

According to reports, several development platforms within Tencent have been connected to the Tencent mixed yuan model, and more than 180 internal businesses have been connected to mixed yuan, including Tencent conference, Tencent documents, WeCom, Tencent Ads and Wechat.

▲ source CTOnews.com netizens recently, some CTOnews.com netizens feedback has passed the internal test review of the mixed meta model, and shared the relevant operation page. As can be seen from the screenshot, the mixed yuan model supports the formulation of interview outline, travel plan, PPT outline, fitness plan and other functions, divided into work, programming, marketing, life and other options.

In addition, the mixed meta model also supports AI dialogue function, which can be tasted by users who have passed the internal test.

CTOnews.com attached more generated images:

▲ Tuyuan goose factory technical official account, the same below

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.