Video Generation Model

On October 10th, IRTS has now released a new generation of video-generated flagship model previews of Vidu Q4 Preview, supporting up to 3 paragraphs reference audio, 15 reference charts and 2K, 4K output. 1AI was informed that Vidu Q4 currently supports Tusheng and Reference Live Videos. Graphical videos can be generated by image and text description; reference videos can be generated by combining multiple reference pictures and audio. Vidu Q4 Preview supports a maximum of 3 paragraphs of reference audio, making role sound more..

On July 18th, Kimi has no plans to develop video-generation models. At this stage, the team concentrates its resources on upgrading the smart ceiling of the model rather than following the competitive pace of video-generated products. Zhou Jin-woo believes that video production, while of clear use, is not very helpful in terms of theoretical and practical effects to enhance model intelligence. He separates capacity enhancement from functional expansion: video generation can add an output form, but it is not the main path in the team's current judgement that drives reasoning and mission capacity to continue to rise..

On June 23rd, according to daily economic news reports, the byte-jumping director of the volcanic engine Tan Tan was officially released today at the 2026 Volcanic Engine FORCE power conference, the soybean video generation model Seedance 2.5. The model is currently in the global enterprise in-house survey phase and is expected to be officially launched in early July. According to a live presentation, Feedance 2.5 achieved a breakthrough in a number of core competencies: supporting the direct generation of single-stage, 30-second videos and allowing users to import up to 50 full-modular reference materials at the same time, significantly increasing video students..

On June 23rd, Ali Baba released video generation model HappyHorse 1.1 yesterday. Officially, the new version 1.0 has been updated in terms of dynamic performance, body consistency, command compliance, visual sense and audio. On June 23rd, Ali Baba released video generation model HappyHorse 1.1 yesterday. Officially, the new version 1.0 has been updated in terms of dynamic performance, body consistency, command compliance, visual sense and audio. Generating specifications, HappyHorse..

News from April 9, yesterday, an anonymous AI video generation model named HappyHorse-1.0 suddenly appeared on the Video Arena list of authoritative AI assessment platforms, Artificial Analysis, and took overboard many tracks. In the Text-Television (No Audio) Track, HappyHorse-1.0 is first in the Elo score of 1332, leading by the second name, Dreamina Seedance 2.0 nearly 60; in..

On March 6, Microsoft announced yesterday (5 March) that its mandatory video creator (Bing Video Creator) had full access to the latest Sora 2 video generation model of OpenAI. Microsoft launched the mandatory video creator service in June 2025, when most of the online AI video generation tools on the market were limited to fee-paying users, while in order to lower the threshold for the use of video generation models, the functionality was made available to all required users free of charge and the system could use the first generation by simply entering text tips..

On December 19, according to a tweet from the LongCat public, the company LongCat team officially released and opened the SOTA-class virtual human video generation model, LongCat-Video-Avatar. The model is based on the LongCat-Video base and continues the core design of "One Model for Multitask" and supports core functions such as Audio-Text-to-Video, Audio-Text-Image-to-Video and video continuation, as well as..

According to news from 28 October, according to Volcanic Engine public tweets, on 24 October, the Volcanic Engine officially went online to generate video-generation model 1.0 beans. Based on the core advantages of the Seedance 1.0pro model, the model achieved a significant efficiency breakthrough: the highest rate of generation increased by about three times, with prices falling by 72%. Seedance 1.0 pro fast achieved a significant increase in generation efficiency, and the 5-second video of 720P could be completed in only 10 seconds, more rapidly than pro version..

October 27th, this morning, the LongCat team released and opened the LongCat-Video Video Generation Model. According to the official presentation, it achieves open-source SOTA (the most advanced level) with a unified model in the context of the Ventian, Graphical video base mission, and is pre-trained in the use of raw video continuation missions, achieves consistent generation of long-minute video, ensures consistency of time series across frames and soundness of physical motion, and has a significant advantage in the area of long video generation. According to the presentation, in recent years, the World Model has allowed artificial intelligence to truly understand..

On October 16th, Google officially released Veo 3.1 Video Generation Model Updates today, with significant improvements at both functional and model levels. In terms of functionality, for the first time, Veo 3.1 added audio support to the "Reflection Video" and "Suspend" functions to make the creation process more complete. Users can not only define roles and styles through multiple reference images, but can also use first-end images to generate seamless transitions, or extend to produce content in excess of one minute. At the model level, Veo 3.1 has made significant progress in the understanding and audio-visual quality of the hints..

On October 1st, in the morning of Beijing time, OpenAI released the next generation video generation model Sora 2. This new model has been significantly enhanced in the definition of video effects and has added an audio generation capability. It inherits and expands early image generation techniques, and in the new version of App launched during the same period, users can complete their identification by recording their own video and voice in one-off, and then “passenger” themselves or others in the resulting video. Sora uses similar to existing social media to provide an algorithm-recommended stream of information and to promote individualized content based on interactive objects and interests..

August 27 news, yesterday evening, Ali Cloud announced the open source of a new multimodal video generation model Tongyi Wan phase Wan2.2-S2V, only a static picture and a piece of audio, you can generate a natural facial expression, consistent mouth shape, body movement silky smooth movie level digital human video. According to reports, the length of the video generated by this model can be up to minute level, which significantly improves the efficiency of video creation in the industries of digital human live broadcasting, movie and TV production, and AI education. Currently, Wan2.2-S2V can drive real people, cartoons, animals, digital people and other types of images, and supports portraits, half-body and full-body any...

July 2 news, Baidu commercial R & D team released its own video generation model "MuseSteamer" and creation platform "Painted Thinking", MuseSteamer is the world's first to realize the integration of Chinese audio and video video generation video model. MuseSteamer is the world's first video model that realizes the integrated generation of Chinese audio and video. The technology can realize the synergistic creation of picture, sound effects and human voice lines, breaking the fragmented process of traditional AIGC video, which is "picture first and then dubbing". It is reported that MuseSteamer tops the world's first in the authoritative list VBench I2V with a total score of 89.38%, and supports the generation of a 10-second 1...

June 10 news, byte jumping volcano engine official public number announced today, will be released on June 11, the new beanbag video generation model. According to the introduction, the new beanbag video generation model has a number of "hard core capabilities", 1AI with examples as follows: support seamless multi-camera narrative, that is, through the efficient model structure, multi-modal positional coding and multi-tasking unified modeling, the model can support unique and stable multi-camera expression. Supports multi-action and on-the-fly camera operation, i.e., after fully learning rich scenes, subjects and behavioral actions, the model can more accurately respond to the user's fine commands, and smoothly generate multi-subject and multi-movement...

Google DeepMind's Veo model, an advanced video generation model with extreme realism and fidelity, support for 4K output, and improved cue-following capabilities and creative control.Veo version 3 adds native audio generation, capable of generating sound effects, ambient soundscapes, and even dialog.Veo is an advanced video generation model that helps creatives make it easier to produce high-quality videos. It supports features such as native audio generation, style matching, and precise camera control for a variety of scenarios, from movie production to game development. v...

May 21 news, in this year's I/O developer conference, Google released a new generation of video generation model Veo3, which is also its first model can generate video background sound effects. It can not only synthesize the picture, but also for the birdsong or street traffic and other scenes with corresponding sound effects, and can even generate character dialog. Google says Veo 3 also excels at physics simulation and lip syncing. Currently, the model is only available to Gemini Ultra users in the U.S., as well as enterprise users of Vertex AI, which has also been integrated into Google's AI Shadow...

May 18 news, B station team May 12 open source anime video generation model AniSora. AniSora can create a variety of anime style video clips with one click, including series episodes, Chinese original animation, manga adaptation, VTuber content, anime PV and ghost animation and so on. The B Station team says that animated content is highly sought after in today's movie and television industry. Although advanced models such as Sora, Kling, and CogVideoX perform well in natural video generation, they are still stretched for anime videos. Moreover, due to...

May 13, 2011 - In the FORCE LINK AI Innovation Tour held in Shanghai today, Volcano Engine officially released a series of AI model upgrades, including Seedance 1.0 lite, a video generation model, and Seedance 1.5, a visual deep-thinking model, as well as upgrades to the Doubao music model, aiming to help enterprises to pass the application chain from business to intelligent body through a more comprehensive model matrix and richer intelligent body tools. With a more comprehensive model matrix and a richer set of intelligent body tools, the company aims to help enterprises open up the application chain from business to intelligent body. According to 1AI, this release of Seedance 1.0 lite...

Another heavyweight open source player has emerged in the video generation space. On April 21, 2025, Sand AI, the startup of Marr Prize and Tsinghua Special Prize winner Cao Yue, launched its own big model for video generation -- MAGI-1. This is a world model that generates videos by autoregressive prediction of video block sequences, with natural and smooth generation effects, and multiple versions available for download. According to the official introduction, the video generated by MAGI-1 has the following characteristics: 1, high smoothness, no lag, can be unlimited renewal. It can be a mirror to the end to generate a continuous long visual...

April 16 - Google has announced that it is bringing its Veo 2 video generation AI model to Gemini Advanced subscribers. The move is designed to counter competition from OpenAI's Sora video generation platform and gain a foothold in an increasingly competitive market. Just two weeks ago, Runway, a strong rival in the synthetic media space, released its fourth-generation video generator and managed to raise more than $300 million (note: roughly Rs. 2,193 million at current exchange rates) in new funding. Starting this Tuesday, Gemi...

April 15 news, fast hand today held "inspiration comes true" can Ling AI 2.0 model conference, and announced that the base model upgraded again, for the world officially released can Ling 2.0 video generation model and can figure 2.0 image generation model. According to the introduction, Ke Ling 2.0 model maintains global leadership in dynamic quality, semantic response, screen aesthetics and other dimensions; Ke Tu 2.0 model significantly improves in command compliance, movie texture and artistic style performance. Gai Kun, senior vice president of Racer and head of the community science line, revealed that since its release in June last year, Ke Ling AI...

April 1, 2011 - Artificial intelligence (AI) startup Runway on Monday released its newly developed AI video generation model, Gen-4, which the company claims is one of the highest fidelity AI-driven video generation tools to date. The newly released Gen-4 model is now available to Runway's personal and enterprise customers. According to Runway, the model's core strengths are its ability to maintain a high degree of consistency of characters, locations, and objects across video scenes, maintaining a "coherent world environment," and the ability to...

March 13 news, today, Lucent Technologies announced the launch of Open-Sora2.0, and fully open source model weights, inference code and distributed training of the entire process. According to reports, this is a new open source SOTA video generation model, only 200,000 U.S. dollars (Note: the current about 1.449 million yuan), that is, 224 GPUs to successfully train a commercial-grade 11B parameters of the video generation of large models, the performance of Tencent Mixed Meta and 30B parameters of the Step-Video. Lucent Technologies said, from Open-Sora ...

Following the Tencent mixed yuan, Ali also announced its open source video generation model: Wan phase 2.1 (Wan2.1), said it has SOTA-level performance, its highlights include: 1, better than the existing open source model, and can even "with some closed source model comparable to. 2, is the first video model that can simultaneously generate Chinese and English text. 3, support for consumer-grade GPUs, of which the T2V-1.3B model requires only 8.19GB of video memory. At present, WanPhase 2.1 can be deployed locally through ComfyUI, the following method: First, the installation of the necessary tools Please keep the "network open"...