Where does AI-generated video go？ ＂Smart AI＂ appeared at Netease Future Conference.
On December 22nd-23rd, Netease Future Conference with the theme of "Intelligent Emergence and Discovery of the Future" was held in Hangzhou, Zhejiang. In the round-table dialogue of "AI Trips" in AGI Forum, Yi Zili, an associate professor at intelligence science and technology College of Nanjing University, Lei Haibo, the founder of "Smart AI", and Nausca, a well-known blogger of AI painting, discussed together on "Where to go for AI-generated video? Warren Wang, partner of Innolux Angel Fund, is the moderator of this dialogue.
Professor Yi Zili first said that with the advent of AI craze, many enterprises have been trying to use AI tools, but the technology of Wensheng map and Wensheng video still needs to be further matured. At present, the main technical path of AI video generation is diffusion model, and the future trend may be to return to the way of large model training. Foreign countries have a certain leading edge in the underlying technology of video generation, and domestic companies have performed better in sub-applications such as 2D digital people and AI social networking. It is believed that with the improvement of computing power level and the innovation of technical paradigm, China may surpass foreign countries in some aspects in the future.
Lei Haibo, founder of "Smart AI"
"Smart AI" is an entrepreneurial enterprise that applies generative AI image technology to marketing design. We have been exploring in the visual design and visual image industry for the past 20 years. Lei Haibo, the founder of "Smart AI", said that in the past, as a design community, media and design platform, he dealt with designers and design agencies almost every day. As far as I know, at present, some top art and design universities in China have applied large models such as Wen Sheng Tu in their daily teaching and design projects. Therefore, there is no doubt about the ability of AI in the field of image generation, but it may take six months to one year to generate video and apply it to the ground.
Facing the problem of why domestic large model manufacturers can’t compete with foreign countries. He bluntly said that the reason for this phenomenon is not only the gap in technology, computing power and data set, but also the lack of aesthetic understanding of domestic manufacturers from the perspective of design. In fact, the tonality, atmosphere, light and shadow texture of the current Midjourney map have far surpassed human performance. If domestic manufacturers can make a visual model similar to Midjourney, and combine high-quality data sets, with industry cognition and industrialization landing ability, there must be great opportunities for vertical application.
When asked about the landing application scenario of "Smart AI", he replied simply and directly, "We mainly face prefix scenarios, such as marketing, especially e-commerce marketing. In the past, the marketing materials of hundreds of millions of merchants and billions of SKUs (single products) were all realized manually. Now imagine, if the uploaded commodity information can not only remain unchanged in the modes of words and images, but also be well integrated with the modal models of words, pictures and videos, so as to produce AI commodity maps, posters, short videos and even 3D interactive content for merchants, this market demand is great. This year, the main force of "Smart AI" is the research and development of Wensheng maps and vertical models and the exploration of industrial applications. However, we have seen the liberation of creative productivity brought by Wensheng maps and graphic videos, and we look forward to the high-quality generation ability of AI in the 3D field. "
As a knowledge blogger and model trainer in the field of AI, Neuska believes that AI can be divided into four types of scenes in the field of video generation: original video style transformation, instant universe, graphic video and scene conversion video. At present, there are many attempts in the fields of advertising production, trailer production, tweet and short video creation. In the aspect of image generation, AI has been able to achieve various dazzling effects, but in the field of video generation, its expressive force is limited due to the lack of control means. For example, the consistency of characters’ expressions in the video is not enough, which is prone to the "horror valley effect". She hopes that the technology can be more accurate in the control of video generation effect. But in the world of AI, creative ideas must be greater than technology.
Warren Wang, Partner of Innolux Angel Fund
From the perspective of investors, Warren Wang said that the popularity of generated video is very high at present. This year, venture capital is mainly concentrated at both ends: first, the capital is concentrated, and only some VCS dare to really shoot; Second, the project is centralized. VC funds are mainly invested in computing power-related projects, such as GPU, chip, high-speed lossless network, large model and so on. Next year, we are looking forward to the field of multimodal models and the application of multimodal capabilities.
Of course, AI, as an intelligent technology, needs the whole society to deepen its understanding, and needs the joint participation of government, scientists, enterprises, media and capital, so as to promote its better and benign development.
It is reported that "Smart AI" has launched nearly 1,000 product map scenes, initially forming an AI tool matrix. Under the control of the subject, the LoRA model with a specific style scene, the special AI model with several hundred million parameters and the intelligent aesthetic evaluation system make the generated product map present unique visual characteristics and aesthetic tonality, thus providing better services for enterprise users.
Reporting/feedback