Is the spring in your heart joyful, sad, calm, or exciting? If you could have an exclusive painting by Cézanne in 5 seconds, which option would you choose?
On July 6, at the 2023 World Artificial Intelligence Conference, upstream news (email: [email protected]) reporters saw that the art poster and virtual image experience generated by AI technology became the most lively area of the exhibition.
Butterfly painting generated by AI technology. Photo by Upstream News Reporter Shi Tingting
Click on the mini program, choose your favorite season, mood and understanding of Cézanne's painting style, and you can generate an exclusive Cézanne painting. This is the artificial intelligence cultural and creative experience function launched by JD Cloud. The staff introduced that behind drawing a "Cézanne famous painting" in one minute is the technical support of JD Cloud AIGC capabilities, which generates a large model based on JD Cloud visual content, and after a lot of training and optimization, it can automatically learn and identify various scenes and data, and generate high-quality paintings with specified styles and content. At present, JD Cloud is integrating visual content generation large model technology with the industry, and trying to land in retail, health, marketing and other scenarios.
In the SenseTime exhibition area, upstream news reporters experienced the "second drawing" program that enters keywords to produce different posters. Enter keywords such as "Chinese style realistic garden" into the system, the system generates different styles of posters in a few seconds, according to needs, you can also add the required materials and elements, even if there is no painting foundation, you can have a poster comparable to professional design, and vivid image.
Using AI technology, you can own a famous painting in 1 minute. Photo by Upstream News Reporter Shi Tingting
The upstream news reporter learned that SenseTime's "SenseMirage" is a creation platform that includes SenseTime's self-developed AIGC large model and convenient LoRA training capabilities, and provides third-party community open source models to accelerate reasoning, providing creators with more convenient and perfect content production and creation tools. The operator optimization technology of flash attention is used in the model, and the graph writing speed is increased by 3 times. Its self-developed large model parameters have been increased to 7 billion, which can bring stronger image generation effects, achieve professional photography-level detail portrayal, and make the generated pictures have super texture and fineness.
Standing in front of the booth, changing cartoon images in seconds, according to their own image, you can also generate AI video anchors, and virtual anchors can also take photos, record videos, live interaction, 24*7 online, how to achieve such magical operations?
Standing in front of a mirror can generate a cartoon image. Photo by Upstream News Reporter Shi Tingting
According to the staff of SenseTime, the "Ruying" platform relied on by the avatar only needs a 2-5 minute of live-action video material shot by mobile phone, which can generate actions, expressions, lip shapes, and even hair strands that can be as natural and realistic as real people, and multi-language proficient digital avatars, of which the similarity of real people is as high as 90%~95%, and the matching degree of lip shape can reach 95%. The platform also opens up the entire process from Wensheng Wen and Wensheng Video, through text-driven, you can freely carry out all kinds of video content production without threshold, at the same time, the platform also provides a wealth of online editing capabilities, including templates, beauty filters, music, backgrounds and foregrounds, etc., can create different works.
In addition, with the improvement of comprehensive performance, the accuracy of virtual humans in English, Japanese, Spanish, Arabic and other languages has increased by more than 30%, making the voice and lip matching of digital humans smoother and more natural; It can also achieve cinematic 4K HD video output. Not only that, in the new version, users can also automatically generate an exclusive digital human image that matches the description by entering prompt words, realize the digital human singing function, conveniently create virtual artists and Internet celebrities, or generate short drama scripts through large models to generate digital human short films.
Upstream News Reporter Shi Tingting
Editor: Zou Yu
Responsible editor: Lang Qingxiang
Review: Guan Yi