In the early morning of January 28, Beijing time, DeepSeek, a large domestic model that recently became popular, released the Janus-Pro multi-modal large model on the GitHub platform and entered the field of Wenshengtu.

DeepSeek said that this big model is an advanced version of the JanusFlow big model released on November 13, 2024.

Compared to the previous generation model, Janus-Pro has optimized training strategies, expanded training data, and the model is larger.

Through these improvements, Janus-Pro has made significant advances in both multi-modal understanding and text-to-image instruction tracking capabilities, while also enhancing the stability of text-to-image generation.

According to test results published by DeepSeek, Janus-Pro beat Stable Diffusion and OpenAI's DALL-E 3 in GenEval and DPG-Bench benchmarks. Currently, 4 models in the Janus series are open source.

The Zhitong Finance App learned that with the launch of Sora, the video creation field ushered in an “iPhone moment”. Judging from the current round of AI application-side development practices, the creative creation field represented by text creation, image creation, and video creativity was widely regarded by the market at the beginning of the AI wave as one of the first industries to be profoundly empowered by AI due to its high fault tolerance rate and high investment.

CITIC Securities released a research report saying that the commercialization of the Wenshengtu and Wensheng video industry is expected to take the lead. In this round of industrial transformation, generative AI technology will be combined with creative tools and software, and will bring two major trends to the creative software industry: first, creative software can benefit from the redistribution of value brought about by generative AI driving cost reduction in the industrial chain; second, the combination of generative AI technology and creative software is expected to lower the threshold of content production and drive the semi-professional creative software market to usher in rapid development.

Wensheng Video Related Concepts Hong Kong Stocks:

Meitu Company (01357): Meitu's business model has been successfully transformed. The business has expanded from the C-side to the B-side productivity scenario. At the same time, it has made full use of AIGC, developed its own AI vision model Miracle Vision, and enabled all Meitu products to promote a continuous increase in subscription penetration.

KUAISHOU-W (01024): Kuaishou's self-developed Wenshengtu large model “can be drawn” The large model currently supports two types of functions: Wenshengtu and Tushengtu, and can be used for AI image creation and AI image customization. This is also the first time that Kuaishou has opened a large series of self-developed models to the public. According to reports, the parameter scale of the “scalable” model reached the level of 1 billion.

Weimeng Group (02013): The company's relevant technical director said that in the past year, the marketing field has been proven to be one of the best application scenarios for AI big model technology, and the short video marketing industry will become one of the most important driving points for Weimeng WAI in 2024.