Shulou(Shulou.com)12/24 Report--

On the occasion of the "Anniversary Celebration" of AIGC, the OpenAI "Palace fight" on this side is non-stop, while the box is staging "the rise of the Planet of the Apes". Within a short period of half a month, from Meta and Adobe to Runway, Stable Diffusion and Wanxing Technology, domestic and foreign technology giants and AI unicorns are speeding up into the AI Vince Video Raceway.

On November 16, Meta launched two new AI video editing features. The first tool, called "Emu Video", automatically generates four-second videos based on subtitles, images, text descriptions, etc., and the other, "Emu Edit", allows users to modify or edit videos more easily through text commands. This tool provides users with a way to delete or replace objects and people from a picture without any professional image editing experience.

On November 22nd, Stability AI released its latest AI model, Stable Video Diffusion, which can generate video from existing images. According to reports, Stable Video Diffusion actually consists of two models-SVD and SVD-XT. SVD can convert still pictures into 14 frames of video, while SVD-XT can increase the number of frames to 24 frames.

On the same day, Runway, an "established" enterprise in the field of AI video generation, also announced its new "motion brush" function. Users only need to smear an area or body in the picture, choose a direction for it, and add exercise intensity, and the motion brush can add controlled motion to the user's generation.

Runway's motion brush function can generate controlled motion

On Nov. 23, Adobe confirmed the acquisition of AI startup Rephrase.ai, which can be converted into videos with user images combined with text scripts and user avatars, which can be used in business scenarios such as marketing, customer communication, and holiday greetings. It is worth noting that this is Adobe's first acquisition in the AI field.

On the evening of November 24, 300624.SZ, a listed company of AIGC software A shares, launched a short video "the Girl's Life" created by AI on its official video number, demonstrating the video generation ability of its multimedia model. Driven by the large model, users only need to input the story outline or clip script to control the starting frame content of the video, and realize the complete generation of the new video content through the large model ability.

Wan Xing's "Skyscreen" big model generates screenshots of the video "the Girl's Life".

Prior to this, Wanxing Technology has predicted the "Skyscreen" big model. According to reports, "Skyscreen" is the first multimedia model in China to focus on video creative applications, which is composed of video models, audio models, picture models, and language models. It has core competencies such as one-click film, AI art design, Wensheng music, audio enhancement, sound analysis, and multilingual dialogue, and continues to iterate on visual, auditory and other multimodal key capabilities.

As a leader in the domestic AIGC field, Wanxing Technology is continuously deepening the AIGC layout, and has launched new AIGC creative products such as Wanxing Ai painting, Wanxing Broadcasting, Wanxing Zhixiu, Wondershare Kwicut, Wondershare VirtuLook, etc.; its Wanxing Meow Ying, Wondershare PDFelement, Wanxing PDF, Yitu Tu, Yitu Tu, Mo Dao overseas version Mockitt and other products have integrated AIGC capabilities, and continue to optimize on the new.

On November 29, AI startup Pika Labs officially launched version 1.0 of the AI video generator. Pika Labs 1.0 uses a brand new AI model that can generate and edit videos in different styles such as 3D animation, animation, cartoons and movies. As soon as the product was launched, it attracted countless AI bosses to "come on", and quickly secured 55 million US dollars in financing, becoming the latest "top stream" of AI video generation.

The latest AI video generates "top stream" Pika Labs

In fact, in this round of AI wave, Wen Sheng Wen, Wen Sheng Tu has been parallel development, ChatGPT represents the first breakthrough in text generation, Midjourney will be available to everyone, and with the gradual extension of the timeline ushered in countless new players. Under the guidance of the "forerunner", the new players obviously took a lot of detours, and the competition between the two tracks has become increasingly fierce.

Different from the direct generation of text and picture, the underlying principle of video is the combination of multi-frame images. Wen Sheng video / picture video needs to increase the time dimension on the basis of text picture. This means that computing costs are high-a short video contains about 30 images per second, and a single video clip has hundreds of thousands of frames, which requires a lot of computing resources to ensure the consistency of space and time between each frame.

In addition, complex information is difficult to deal with-video has visual dynamic information, and it becomes very challenging to model video content after adding time information between different frames.

Finally, the high-intensity requirement of algorithm stability-- the high requirement of video coherence, means that the AI generation of each frame needs high correlation and continuous and smooth changes.

Everything is one and two sides. AI video generation technology is still in rapid iteration and rapid evolution, and how many new opportunities it will give birth to is still unknown. The only thing that is certain is that the players who have entered the game have obviously grabbed the opportunity to compete for the best.

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.