AI model

Kling2.5 Turbo is an AI video generation model that significantly improves the understanding of complex causal relationships and time series. It has the characteristics of cost-optimized generation. The cost of generating a 5-second high-quality video is reduced by 30% (25 points vs. 35 points), and the motion smoothness is excellent. It uses advanced reasoning intelligence to understand complex causal relationships and time instructions, greatly improving motion smoothness and camera stability while optimizing costs. It's also the world's first model to output native 10, 12 and 16-bit HDR video in EXR format, suitable for professional studio workflows and pipelines. Additionally, its draft mode generates 20 times faster, making it easy to iterate quickly. The product has a variety of price plans, including a free entry version, a $29 professional version, and a $99 studio version, suitable for users with different needs, from individual creators to corporate teams.

iMideo is an AI video generation platform with multiple advanced AI models such as Veo3 and Seedance. Its main advantage is that it can quickly convert still pictures into high-quality AI videos without complex editing skills, and it supports multiple aspect ratios and resolution settings. The platform provides a free version, allowing users to try the image-to-video function for free first. The paid plan starts at US$5.95 per month, which is suitable for all types of creators to easily produce professional-level video content.

Ray 3 is the first video AI inference model launched by Lumakey, capable of generating true EXR 10, 12, 12, 12 and 16-bit HDR format videos. Its importance lies in providing new tools for high-quality video production to the film, television and advertising industries. The main advantages include high-bit HDR format, with better color and brightness performance, suitable for high-end projects; it can be used for high-resolution video production to meet professional needs. The product background is to meet the demand for high-quality videos in the film, television and advertising industries. Regarding the price, the documentation does not mention it. Product positioning is to serve the fields of high-end film and television and advertising production.

Ray3 is the world's first video model with inference capabilities, powered by Luma Ray3. It can think, plan and create professional-grade content, with native HDR generation and intelligent draft mode for rapid iteration. Key benefits include: inferential intelligence to deeply understand prompts, plan complex scenes, and self-examine; native 10, 12, and 16-bit HDR video for professional studio workflows; and draft mode to generate 20 times faster, making it easy to refine concepts quickly. In terms of price, there is a free version, a $29 professional version and a $99 studio version. Positioned to meet the video creation needs of different user groups from exploration to professional commercial applications.

Ray3 is the world's first AI video model with inference intelligence and 16-bit HDR output. Its importance lies in providing advanced video generation solutions for film and television producers, advertising companies and studios. Its main advantages are: the output video has high fidelity, consistency and controllability; it supports 16-bit HDR, providing professional-level color depth and dynamic range; it has reasoning intelligence and can understand the scene context to ensure the logical consistency and physical accuracy of each frame; it is compatible with Adobe software and can be seamlessly integrated into the existing production process; it has a 5x speed draft mode for rapid creative testing. This product is positioned in the field of professional video production. Although the specific price is not mentioned in the document, there is a "trial" option, and it is speculated that it may adopt a free trial plus payment model.

Lucy Edit AI is the first basic model for text-guided video editing, launched by DecartAI and open source. Its importance lies in innovating the video creation model, allowing creators to edit videos only through text commands without complicated operations. Key benefits include lightning-fast processing speeds, industry-leading accuracy, unlimited video creation potential, a simple and intuitive interface, and is trusted by content creators around the world. This product is free to use and is positioned to help users complete professional video editing efficiently and conveniently.

Ray 3 AI Video Generator is a video generation platform driven by advanced Ray 3 AI technology. It is the world's first AI video model with HDR generation and intelligent reasoning capabilities. Its importance lies in providing professional creators and enterprises with powerful video production tools that can quickly convert text into high-quality 4K HDR videos. The main advantages include intelligent reasoning to understand user intentions, support for multiple video styles, and multiple practical functions such as voice narration, smart subtitles, etc. The product background was developed to meet the market's demand for efficient, high-quality video creation. In terms of price, there is a free version, a professional version ($29.9 per month) and an enterprise version ($999). It is positioned to serve creators and enterprises around the world and assist professional HDR video creation.

Hailuo 2 is an AI video generator that uses MoE technology to convert text and images into 720P videos. Its main advantages include advanced AI technology, high-definition video generation, text-to-video function, etc.

Wan 2.2 is an AI video generator that uses advanced MoE technology to convert text and images into 720P videos. It supports consumer-grade GPUs and can generate professional videos in real time.

Veo 5 AI Video Generator is a next-generation AI video generator based on Veo 5 technology that can quickly create stunning, ultra-realistic videos. It uses the latest Veo 5 A model to achieve intelligent scene understanding, natural motion synthesis and context-aware rendering, bringing unprecedented ultra-realism and creativity.

LTXV 13B is an advanced AI video generation model developed by Lightricks with 13 billion parameters, significantly improving the quality and speed of video generation. Released in May 2025, this model is a significant upgrade from its predecessor, the LTX video model, supporting real-time high-quality video generation and suitable for all types of creative content production. The model uses multi-scale rendering technology to generate 30 times faster than similar models and run smoothly on consumer hardware.

Veo3 AI Video Generator is a powerful tool that uses Google's Veo3 AI model to generate stunning 4K videos from text. Featuring advanced physics simulations and realistic visual effects, transform your ideas into cinematic content. Price: Paid.

Seedance AI is a powerful video model that can generate high-quality, narrative videos from simple text prompts. It has features such as dynamic lens movement and 1080p high-definition video output, providing users with the convenience of creating movie-level videos.

DreamASMR leverages Veo3 ASMR technology to create relaxing video content, providing advanced AI video generation, binaural sound and a meticulous visual experience, making it the ultimate ASMR experience.

LIP Sync AI is a revolutionary AI technology that utilizes a global audio perception engine to transform still photos into lifelike conversational videos. Its main advantage is its efficient and realistic generation of effects, resulting in photos with perfect lip synchronization. This product is positioned to provide users with high-quality lip sync video generation services.

Veo3 Video is a platform that uses the Google Veo3 model to generate high-quality videos. It uses advanced technology and algorithms to ensure audio and lip synchronization during video generation, providing consistent video quality.

Veo 3 is the latest video generation model designed to deliver 4K output with greater realism and audio effects that more accurately follow user cues. This technology represents a major advancement in video generation, allowing for greater creative control. The launch of Veo 3 is a major upgrade to Veo 2 and is designed to help creators realize their creative visions. This product is suitable for creative industries that require high-quality video generation, ranging from advertising to game development. No specific price information was disclosed.

Index-AniSora is a top animated video generation model open sourced by Bilibili. It is implemented based on AniSora technology and supports one-click generation of a variety of two-dimensional style video shots, such as Fanju, Guochuang, comic animation, VTuber, animated PV and ghost animation, etc. This model comprehensively improves the production efficiency and quality of animation content through the reinforcement learning technology framework, and its technical principles have been accepted by IJCAI2025. The open source of Index-AniSora has brought new technological breakthroughs to the field of animated video generation, providing developers and creators with powerful tools to promote the further development of two-dimensional content creation.

HunyuanCustom is a multi-modal custom video generation framework designed to generate topic-specific videos based on user-defined conditions. This technology performs well in identity consistency and supports multiple input modes. It can handle text, image, audio and video input, and is suitable for a variety of application scenarios such as virtual human advertising and video editing.

KeySync is a leak-free lip-syncing framework for high-resolution video. It solves the temporal consistency problem in traditional lip sync technology while handling expression leakage and facial occlusion through clever masking strategies. The superiority of KeySync is reflected in its advanced results in lip reconstruction and cross-synchronization, which is suitable for practical application scenarios such as automatic dubbing.

Vidu Q1 is a large domestic video generation model launched by Shengshu Technology. It is specially designed for video creators and supports high-definition 1080p video generation, with movie-level camera effects and first and last frame functions. The product topped the list in both VBench-1.0 and VBench-2.0 reviews and is extremely cost-effective at only one-tenth the price of its peers. It is suitable for many fields such as movies, advertising, animation, etc., and can significantly reduce creative costs and improve creative efficiency.

SkyReels-V2 is the world's first infinite-duration movie generation model using a diffusion forcing framework released by the Kunlun Wanwei SkyReels team. This model achieves collaborative optimization by combining multi-modal large language models, multi-stage pre-training, reinforcement learning and diffusion forcing frameworks, breaking through the major challenges of traditional video generation technology in prompt word compliance, visual quality, motion dynamics and video duration coordination. Not only does it provide powerful tools for content creators, it also opens up endless possibilities for using AI for video storytelling and creative expression.

Wan2.1-FLF2V-14B is an open source large-scale video generation model designed to advance the field of video generation. The model performs well in multiple benchmark tests, supports consumer-grade GPUs, and can efficiently generate 480P and 720P videos. It performs well in multiple tasks such as text to video and image to video. It has powerful visual text generation capabilities and is suitable for various practical application scenarios.

FramePack is an innovative video generation model designed to improve the quality and efficiency of video generation by compressing the context of input frames. Its main advantage is that it solves the drift problem in video generation and maintains video quality through a bidirectional sampling method, making it suitable for users who need to generate long videos. The technical background comes from in-depth research and experiments on existing models to improve the stability and coherence of video generation.

Pusa introduces an innovative method of video diffusion modeling through frame-level noise control, which enables high-quality video generation and is suitable for a variety of video generation tasks (text to video, image to video, etc.). With its excellent motion fidelity and efficient training process, this model provides an open source solution to facilitate users in video generation tasks.

SkyReels-A2 is a video diffusion transformer-based framework that allows users to synthesize and generate video content. This model provides flexible creative capabilities by leveraging deep learning technology and is suitable for a variety of video generation applications, especially in animation and special effects production. The advantage of this product is its open source nature and efficient model performance, which is suitable for researchers and developers and is currently free of charge.

DreamActor-M1 is a Diffusion Transformer (DiT)-based human animation framework designed to achieve fine-grained global controllability, multi-scale adaptability, and long-term temporal consistency. Through hybrid guidance, the model is able to generate highly expressive and photorealistic human videos, suitable for a variety of scenarios from portraits to full-body animations. Its main advantages are high fidelity and identity preservation, bringing new possibilities for animation of human behavior.

MoCha is an innovative technology designed to synthesize high-quality dialogue characters, making it widely applicable in film and television production, games and animation. The main advantage of this technology is that it can generate more natural and smooth character dialogue, which enhances the audience's immersion. MoCha's market positioning is for professional film and television production companies and independent developers, committed to improving the realism of character interaction. The product adopts a model based on deep learning, the price strategy is paid, and different levels of service packages are provided.

GAIA-2 is an advanced video generation model developed by Wayve, designed to provide diverse and complex driving scenarios for autonomous driving systems to improve safety and reliability. The model addresses the limitations of relying on real-world data collection by generating synthetic data, capable of creating a variety of driving scenarios, including routine and edge cases. GAIA-2 supports simulation of a variety of geographical and environmental conditions, helping developers quickly test and verify autonomous driving algorithms without high costs.

AccVideo is a novel and efficient distillation method that accelerates the inference of video diffusion models using synthetic datasets. The model is able to achieve an 8.5x speedup in generating videos while maintaining similar performance. It uses a pre-trained video diffusion model to generate multiple effective denoised trajectories, thus optimizing the data usage and generation process. AccVideo is particularly suitable for scenarios that require efficient video generation, such as film production, game development, etc., and is suitable for researchers and developers.

Video-T1 is a video generation model that significantly improves the quality and consistency of generated videos through test time scaling technology (TTS). This technology allows the use of more computing resources during inference, thus optimizing the generated results. Compared with traditional video generation methods, TTS can provide higher generation quality and richer content expression, and is suitable for the field of digital creation. The product is positioned primarily for researchers and developers, and pricing information is not clear.

Step-Video-TI2V is an advanced graphic video model developed by Shanghai Step Star Intelligent Technology Co., Ltd. It is trained based on Step-Video-T2V with 30B parameters and can generate videos up to 102 frames based on text and image input. The core advantage of this model lies in its two major features: controllable motion amplitude and controllable lens motion, which can balance the dynamics and stability of video generation results. In addition, it performs well in animation-style video generation and is very suitable for application scenarios such as animation creation and short video production. The open source of this model provides powerful technical support for the field of video generation and promotes the development of multi-modal generation technology.

SmolVLM2 is a lightweight video language model designed to generate relevant text descriptions or video highlights by analyzing video content. This model is efficient, has low resource consumption, and is suitable for running on a variety of devices, including mobile devices and desktop clients. Its main advantage is that it can quickly process video data and generate high-quality text output, providing powerful technical support for video content creation, video analysis, education and other fields. This model was developed by the Hugging Face team and is positioned as an efficient and lightweight video processing tool. It is currently in the experimental stage and users can try it for free.

HunyuanVideo-I2V is Tencent's open source image-to-video generation model, developed based on the HunyuanVideo architecture. This model effectively integrates reference image information into the video generation process through image latent stitching technology, supports high-resolution video generation, and provides customizable LoRA effect training functions. This technology is of great significance in the field of video creation, as it can help creators quickly generate high-quality video content and improve creation efficiency.

Project Starlight is an AI video enhancement model from Topaz Labs designed to improve the quality of low-resolution and corrupted videos. It uses diffusion model technology to achieve video super-resolution, noise reduction, deblurring, and sharpening functions while maintaining temporal consistency and ensuring smooth transitions between video frames. This technology is a major breakthrough in the field of video enhancement, bringing unprecedented high-quality effects to video repair and enhancement. Currently, Project Starlight offers a free trial, with plans to support 4K export in the future, primarily for users and businesses in need of high-quality video restoration and enhancement.

PSYCHE AI is a tool focused on generating realistic AI videos. Its core function is to quickly generate high-quality video content through AI technology. Users can choose from over 100 AI characters and 120 AI voices to generate content without any video editing experience. This product is based on advanced AI technology and can provide efficient content creation solutions for enterprises and individuals, especially suitable for areas such as content marketing, education, digital employees and personalized brands. Its price is positioned at US$2-3 per video, which is significantly lower than traditional video production costs. It also provides a free trial, which lowers the user threshold.

Wan2GP is an improved version based on Wan2.1, designed to provide efficient, low-memory video generation solutions for low-configuration GPU users. This model enables ordinary users to quickly generate high-quality video content on consumer-grade GPUs by optimizing memory management and acceleration algorithms. It supports a variety of tasks, including text to video, image to video, video editing, etc., and has a powerful video VAE architecture that can efficiently process 1080P videos. The emergence of Wan2GP has lowered the threshold of video generation technology, allowing more users to easily get started and apply it to actual scenarios.

HunyuanVideo Keyframe Control Lora is an adapter for the HunyuanVideo T2V model, focusing on keyframe video generation. It achieves efficient fine-tuning by modifying the input embedding layer to effectively integrate keyframe information, and applying low-rank adaptation (LoRA) technology to optimize linear layers and convolutional input layers. This model allows users to precisely control the starting and ending frames of the generated video by defining key frames, ensuring that the generated content is seamlessly connected to the specified key frames, enhancing video coherence and narrative. It has important application value in the field of video generation, especially in scenarios where precise control of video content is required.

Wan2.1 is an open source, advanced large-scale video generation model designed to push the boundaries of video generation technology. It significantly improves model performance and versatility through innovative spatiotemporal variational autoencoders (VAE), scalable training strategies, large-scale data construction, and automated evaluation metrics. Wan2.1 supports a variety of tasks, including text to video, image to video, video editing, etc., and is capable of generating high-quality video content. The model performs well on multiple benchmarks, even surpassing some closed-source models. Its open source nature allows researchers and developers to freely use and extend the model for a variety of application scenarios.

Wan2.1-T2V-14B is an advanced text-to-video generation model based on a diffusion transformer architecture that combines an innovative spatiotemporal variational autoencoder (VAE) with large-scale data training. It is capable of generating high-quality video content at multiple resolutions, supports Chinese and English text input, and surpasses existing open source and commercial models in performance and efficiency. This model is suitable for scenarios that require efficient video generation, such as content creation, advertising production, and video editing. The model is currently available for free on the Hugging Face platform and is designed to promote the development and application of video generation technology.

SkyReels V1 is a human-centered video generation model fine-tuned based on HunyuanVideo. It is trained through high-quality film and television clips to generate video content with movie-like quality. This model has reached the industry-leading level in the open source field, especially in facial expression capture and scene understanding. Its key benefits include open source leadership, advanced facial animation technology and cinematic light and shadow aesthetics. This model is suitable for scenarios that require high-quality video generation, such as film and television production, advertising creation, etc., and has broad application prospects.

SkyReels-V1 is an open source human-centered video basic model, fine-tuned based on high-quality film and television clips, focusing on generating high-quality video content. This model has reached the top level in the open source field and is comparable to commercial models. Its main advantages include: high-quality facial expression capture, cinematic light and shadow effects, and the efficient inference framework SkyReelsInfer, which supports multi-GPU parallel processing. This model is suitable for scenarios that require high-quality video generation, such as film and television production, advertising creation, etc.

FlashVideo is a deep learning model focused on efficient high-resolution video generation. It uses a staged generation strategy to first generate low-resolution videos and then upgrade them to high resolutions through enhanced models, thereby significantly reducing computational costs while ensuring details. This technology is of great significance in the field of video generation, especially in scenarios where high-quality visual content is required. FlashVideo is suitable for a variety of application scenarios, including content creation, advertising production, and video editing. Its open source nature allows researchers and developers the flexibility to customize and extend it.

Light-A-Video is an innovative video relighting technology designed to solve the lighting inconsistency and flicker issues present in traditional video relighting. This technology enhances lighting consistency between video frames while maintaining high-quality image effects through the Consistent Light Attention (CLA) module and Progressive Light Fusion (PLF) strategy. This technology requires no additional training and can be directly applied to existing video content, making it efficient and practical. It is suitable for video editing, film and television production and other fields, and can significantly improve the visual effect of videos.

Magic 1-For-1 is a model focused on efficient video generation. Its core function is to quickly convert text and images into videos. This model optimizes memory usage and reduces inference latency by decomposing the text-to-video generation task into two sub-tasks: text-to-image and image-to-video. Its main advantages include efficiency, low latency, and scalability. This model was developed by the Peking University DA-Group team to promote the development of interactive basic video generation. Currently, the model and related code are open source and users can use it for free, but they must abide by the open source license agreement.

LipSync Studio is a professional tool focused on video lip synchronization, using advanced artificial intelligence technology to achieve a perfect match between audio and video. It automatically analyzes and maps mouth movements to ensure every syllable, pause and expression is perfectly aligned with the audio track. This product supports multiple languages ​​and is suitable for video localization, dubbing, comedy creation and other scenarios. It can help content creators quickly generate high-quality multi-lingual video content and improve the global dissemination efficiency of content. Its main advantages include efficient and accurate lip synchronization, as well as powerful multi-language support and batch processing capabilities. The product is positioned to provide powerful tool support for professional video producers, educators, corporate marketers, and social media creators.

On-device Sora is an open source project that aims to achieve efficient video generation on mobile devices such as iPhone 15 Pro through technologies such as Linear Scale Leaping (LPL), Temporal Dimension Tag Merging (TDTM), and Dynamic Loading Concurrent Reasoning (CI-DL). The project is developed based on the Open-Sora model and is capable of generating high-quality videos based on text input. Its main advantages include high efficiency, low power consumption and optimization for mobile devices. This technology is suitable for scenarios where video content needs to be quickly generated on mobile devices, such as short video creation, advertising production, etc. The project is currently open source and users can use it for free.

Lumina-Video is a video generation model developed by the Alpha-VLLM team, mainly used to generate high-quality video content from text. This model is based on deep learning technology and can generate corresponding videos based on text prompts input by users, which is efficient and flexible. It is of great significance in the field of video generation, providing content creators with powerful tools to quickly generate video materials. The project is currently open source, supports video generation at multiple resolutions and frame rates, and provides detailed installation and usage guides.

Goku is an artificial intelligence model focused on video generation, capable of generating high-quality video content based on text prompts. The model is based on advanced streaming generation technology and is capable of generating smooth and attractive videos, suitable for a variety of scenarios such as advertising, entertainment, and creative content production. The main advantage of Goku lies in its efficient generation capabilities and excellent performance of complex scenes, which can significantly reduce video production costs while increasing the attractiveness of content. The model was jointly developed by research teams from the University of Hong Kong and ByteDance to advance the development of video generation technology.

VideoWorld is a deep generative model focused on learning complex knowledge from purely visual input (unlabeled videos). It uses autoregressive video generation technology to explore how to learn task rules, reasoning and planning capabilities through visual information only. The core advantage of this model lies in its innovative latent dynamic model (LDM), which can efficiently represent multi-step visual changes, thereby significantly improving learning efficiency and knowledge acquisition capabilities. VideoWorld performed well in video Go and robot control tasks, demonstrating its strong generalization capabilities and learning capabilities for complex tasks. The research background of this model stems from the imitation of organisms learning knowledge through vision rather than language, and aims to open up new ways for artificial intelligence to acquire knowledge.

VideoJAM is an innovative video generation framework designed to improve motion coherence and visual quality of video generation models through joint appearance-motion representation. This technology introduces an internal guidance mechanism (Inner-Guidance) and uses the motion signals predicted by the model itself to dynamically guide video generation, thus performing well in generating complex motion types. The main advantage of VideoJAM is its ability to significantly improve the coherence of video generation while maintaining high-quality visuals, and can be applied to any video generation model without requiring large-scale modifications to the training data or model architecture. This technology has important application prospects in the field of video generation, especially in scenes that require a high degree of motion coherence.

Go with the Flow is an innovative video generation technology that achieves efficient control of motion patterns in video diffusion models by using distortion noise instead of traditional Gaussian noise. This technology can achieve precise control of object and camera motion in videos without modifying the original model architecture without increasing computational costs. Its main advantages include efficiency, flexibility and scalability, and it can be widely used in various scenarios such as image to video generation and text to video generation. This technology was developed by researchers from institutions such as Netflix Eyeline Studios. It has high academic value and commercial application potential. It is currently open source and freely available to the public.

OmniHuman-1 is an end-to-end multi-modal conditional human video generation framework capable of generating human videos based on a single human image and motion signals (such as audio, video, or a combination thereof). This technology overcomes the problem of scarcity of high-quality data through a hybrid training strategy, supports image input with any aspect ratio, and generates realistic human videos. It performs well in weak signal input (especially audio) and is suitable for a variety of scenarios, such as virtual anchors, video production, etc.

MatAnyone is an advanced video keying technology focused on achieving stable video keying through consistent memory propagation. It uses a region-adaptive memory fusion module and combines target-specified segmentation maps to maintain semantic stability and detail integrity in complex backgrounds. The importance of this technology lies in its ability to provide high-quality keying solutions for video editing, special effects production and content creation, especially for scenes that require precise keying. The main advantages of MatAnyone are its semantic stability in core regions and fine processing of boundary details. It was developed by a research team from Nanyang Technological University and SenseTime to solve the shortcomings of traditional keying methods in complex backgrounds.

Video Depth Anything is a deep learning-based video depth estimation model that provides high-quality, time-consistent depth estimation for extremely long videos. This technology is developed based on Depth Anything V2 and has strong generalization capabilities and stability. Its main advantages include depth estimation capabilities for videos of arbitrary length, temporal consistency, and good adaptability to open-world videos. This model was developed by ByteDance’s research team to solve challenges in depth estimation in long videos, such as temporal consistency issues and adaptability issues in complex scenes. Currently, the code and demonstration of the model are publicly available for researchers and developers to use.

Tarsier is a series of large-scale video language models developed by the ByteDance research team, designed to generate high-quality video descriptions and have powerful video understanding capabilities. This model significantly improves the accuracy and detail of video description through a two-stage training strategy (multi-task pre-training and multi-granularity instruction fine-tuning). Its main advantages include high-precision video description capabilities, the ability to understand complex video content, and SOTA (State-of-the-Art) results in multiple video understanding benchmarks. Tarsier's background is based on improving the shortcomings of existing video language models in description details and accuracy. Through large-scale high-quality data training and innovative training methods, it has reached new heights in the field of video description. This model currently has no clear pricing. It is mainly aimed at academic research and commercial applications, and is suitable for scenarios that require high-quality video content understanding and generation.

leapfusion-hunyuan-image2video is an image-to-video generation technology based on the Hunyuan model. It uses advanced deep learning algorithms to convert static images into dynamic videos, providing content creators with a new way of creation. Key benefits of this technology include efficient content generation, flexible customization capabilities, and support for high-quality video output. It is suitable for scenarios where video content needs to be generated quickly, such as advertising production, video special effects and other fields. The model is currently released as open source for free use by developers and researchers, and its performance is expected to be further improved through community contributions in the future.

VideoLLaMA3 is a cutting-edge multi-modal basic model developed by the DAMO-NLP-SG team, focusing on image and video understanding. The model is based on the Qwen2.5 architecture and combines advanced visual encoders (such as SigLip) and powerful language generation capabilities to handle complex visual and language tasks. Its main advantages include efficient spatiotemporal modeling capabilities, powerful multi-modal fusion capabilities, and optimized training on large-scale data. This model is suitable for application scenarios that require deep video understanding, such as video content analysis, visual question answering, etc., and has extensive research and commercial application potential.

Seaweed-APT is a model for video generation that achieves large-scale text-to-video single-step generation through adversarial post-training techniques. This model can generate high-quality videos in a short time, which has important technical significance and application value. Its main advantages are fast speed and good generation effect, and it is suitable for scenarios where video needs to be generated quickly. The specific price and market positioning have not yet been determined.

Luma Ray2 is an advanced video generation model trained on Luma's new multi-modal architecture with 10 times the computing power of Ray1. It understands text commands and accepts image and video input to generate videos with fast, coherent motion, ultra-realistic detail, and logical sequence of events, bringing the resulting video closer to a production-ready state. Text-to-video generation is currently available, with image-to-video, video-to-video and editing functions coming soon. The product is mainly aimed at users who need high-quality video generation, such as video creators, advertising companies, etc. It is currently only open to paying subscribers and can be tried through the official website link.

The multi-modal model developed by the ictnlp team uses only one visual token to improve performance and improve multiple properties. It is open source and free, and is suitable for scenarios that require quick and accurate understanding of visual content.

Diffusion as Shader (DaS) is an innovative video generation control model designed to achieve diversified control of video generation through the diffusion process of 3D perception. This model utilizes 3D tracking video as control input and can support multiple video control tasks under a unified architecture, such as mesh-to-video generation, camera control, motion transfer, and object manipulation. The main advantage of DaS is its 3D perception capability, which can effectively improve the temporal consistency of generated videos and demonstrate powerful control capabilities through fine-tuning with a small amount of data in a short time. This model was jointly developed by research teams from many universities including the Hong Kong University of Science and Technology. It aims to promote the development of video generation technology and provide more flexible and efficient solutions for film and television production, virtual reality and other fields.

TransPixar is an advanced text-to-video generative model capable of generating RGBA videos that include a transparency channel. The technology achieves highly consistent generation of RGB and Alpha channels by combining a Diffusion Transformer (DiT) architecture and LoRA’s fine-tuning-based approach. TransPixar has important application value in the fields of visual effects (VFX) and interactive content creation, and can provide diverse content generation solutions for industries such as entertainment, advertising, and education. Its main advantages include efficient model scalability, powerful generation capabilities, and optimized processing capabilities for limited training data.

STAR is an innovative video super-resolution technology that solves the over-smoothing problem existing in traditional GAN ​​methods by combining a text-to-video diffusion model with video super-resolution. This technology can not only restore the details of the video, but also maintain the spatiotemporal consistency of the video, making it suitable for various real-world video scenarios. STAR was jointly developed by Nanjing University, ByteDance and other institutions and has high academic value and application prospects.

SeedVR is an innovative diffusion transformer model specifically designed to handle real-world video inpainting tasks. The model is able to efficiently process video sequences of arbitrary length and resolution through its unique shifted window attention mechanism. SeedVR is designed to achieve significant improvements in both generative power and sampling efficiency, performing well on both synthetic and real-world benchmarks compared to traditional diffusion models. In addition, SeedVR incorporates modern practices such as causal video autoencoders, hybrid image and video training, and progressive training, further improving its competitiveness in the field of video restoration. As a cutting-edge video restoration technology, SeedVR provides video content creators and post-production staff with a powerful tool that can significantly improve video quality, especially when working with low-quality or damaged video footage.

LatentSync is a lip sync framework developed by ByteDance based on the latent diffusion model of audio conditions. It directly leverages the power of Stable Diffusion to model complex audio-visual correlations without any intermediate motion representation. This framework effectively improves the temporal consistency of generated video frames while maintaining the accuracy of lip synchronization through the proposed temporal representation alignment (TREPA) technology. This technology has important application value in fields such as video production, virtual anchoring, and animation production. It can significantly improve production efficiency, reduce labor costs, and bring users a more realistic and natural audio-visual experience. The open source nature of LatentSync also enables it to be widely used in academic research and industrial practice, promoting the development and innovation of related technologies.

This is a video variational autoencoder (VAE) designed to reduce video redundancy and promote efficient video generation. This model observes that extending image VAE directly to 3D VAE will introduce motion blur and detail distortion, so time-aware spatial compression is proposed to better encode and decode spatial information. Additionally, the model integrates a lightweight motion compression model to achieve further temporal compression. By leveraging the inherent text information in text-to-video datasets and incorporating text guidance into the model, the reconstruction quality is significantly improved, especially in terms of detail preservation and temporal stability. The model also improves its generality by jointly training on images and videos, which not only improves the reconstruction quality but also enables the model to perform autoencoding of images and videos. Extensive evaluation shows that the method outperforms recent strong baselines.

Video Prediction Policy (VPP) is a robotic policy based on Video Diffusion Models (VDMs) that accurately predicts future image sequences, demonstrating a good understanding of physical dynamics. VPP utilizes visual representations in VDMs to reflect the evolution of the physical world, and this representation is called predictive visual representation. By combining diverse human or robot manipulation datasets and employing a unified video generation training objective, VPP outperforms existing methods in two simulated environments and two real-world benchmarks. Particularly on the Calvin ABC-D benchmark, VPP achieved a 28.1% relative improvement over the previous best technology and a 28.8% improvement in success rate in complex real-world dexterous hand manipulation tasks.

Ruyi-Mini-7B is an open source image-to-video generation model developed by the CreateAI team. It has about 7.1 billion parameters and is capable of generating video frames in 360p to 720p resolution from input images, up to 5 seconds long. Models support different aspect ratios and have enhanced motion and camera controls for greater flexibility and creativity. The model is released under the Apache 2.0 license, which means users can freely use and modify it.

Enhance-A-Video is a project dedicated to improving the quality of video generation by adjusting the temporal attention parameters in the video model to enhance the consistency and visual quality between video frames. The project was developed by researchers from the National University of Singapore, Shanghai Artificial Intelligence Laboratory and the University of Texas at Austin. The main advantage of Enhance-A-Video is that it can improve the performance of existing video models at zero cost and without the need for retraining. It controls inter-frame correlation by introducing temperature parameters to enhance the temporal attention output of the video, thereby improving video quality.

Ruyi is a large Tucson video model released by TuSimple. It is specially designed to run on consumer-grade graphics cards and provides detailed deployment instructions and ComfyUI workflow so that users can get started quickly. Ruyi will provide new possibilities for visual storytelling with its excellent performance in frame-to-frame consistency, smoothness of movements, and harmonious and natural color presentation and composition. At the same time, this model also performs deep learning for animation and game scenes, and will become an ideal creative partner for ACG enthusiasts.

FastHunyuan is an accelerated version of the HunyuanVideo model developed by Hao AI Lab. It can generate high-quality videos in 6 diffusion steps. Compared with the 50-step diffusion of the original HunyuanVideo model, the speed is increased by about 8 times. This model is trained on the MixKit data set for consistent distillation. It has the characteristics of high efficiency and high quality, and is suitable for scenarios that require rapid video generation.

This is a video invisible object segmentation and content completion model proposed by Carnegie Mellon University. This model uses the basic knowledge of video generation models to process visible object sequences in videos through conditional generation tasks to generate object masks and RGB content including visible and invisible parts. The main advantages of this technology include the ability to handle highly occlusion situations and the ability to effectively handle deformed objects. In addition, the model outperforms existing advanced methods on multiple data sets, especially in non-visible segmentation of occluded areas of objects, with performance improvements of up to 13%.

Apollo is an advanced family of large-scale multi-modal models focused on video understanding. It provides practical insights into optimizing model performance by systematically exploring the design space of video-LMMs, revealing the key factors that drive performance. By discovering 'Scaling Consistency', Apollo enables design decisions on smaller models and data sets to be reliably transferred to larger models, significantly reducing computational costs. Apollo's key benefits include efficient design decisions, optimized training plans and data blending, and a new benchmark, ApolloBench, for efficient evaluation.

Veo 2 is the latest video generation model developed by Google DeepMind, which represents a major advancement in video generation technology. Veo 2 is able to realistically simulate real-world physics and a wide range of visual styles while following simple and complex instructions. The model significantly outperforms other AI video models in terms of detail, realism, and reduced artifacts. Veo 2’s advanced motion capabilities allow it to accurately represent motion and follow detailed instructions to create a variety of shot styles, angles and movements. The importance of Veo 2 in the field of video generation is reflected in its enhanced diversity and quality of video content, providing powerful technical support for film production, game development, virtual reality and other fields.

CausVid is an advanced video generation model that enables instant video frame generation by adapting a pre-trained bidirectional diffusion transformer into a causal transformer. The importance of this technology is that it significantly reduces the latency of video generation, allowing video generation to be streamed on a single GPU at an interactive frame rate (9.4FPS). The CausVid model supports text-to-video generation and zero-sample image-to-video generation, demonstrating a new level of video generation technology.

SynCamMaster is an advanced video generation technology that can simultaneously generate multi-camera video from diverse viewpoints. This technology enhances the dynamic consistency of video content under different viewing angles through pre-trained text-to-video models, which is of great significance for application scenarios such as virtual shooting. The main advantages of this technology include the ability to handle arbitrary perspective generation of open-world videos, integrating 6 degrees of freedom camera poses, and designing a progressive training scheme that uses multi-camera images and monocular videos as supplements to significantly improve model performance.

EndlessAI is a platform with AI video capabilities as its core and is currently in stealth mode. It is available as a demo on the App Store through the Lloyd smartphone app, through which users can experience the power of AI video technology. EndlessAI's technical background emphasizes its professionalism in video processing and AI applications. Although the price and specific positioning information are not clear on the page, it can be speculated that it is mainly targeted at user groups who require high-end video processing and AI integrated solutions.

MEMO is an advanced open-weight model for audio-driven speaking video generation. The model enhances long-term identity consistency and motion smoothness through a memory-guided temporal module and an emotion-aware audio module, while refining facial expressions by detecting emotions in audio to generate identity-consistent and expressive speaking videos. Key benefits of MEMO include more realistic video generation, better audio-lip sync, identity consistency and expression emotion alignment. The technical background information shows that MEMO generates more realistic speaking videos across multiple image and audio types, surpassing existing state-of-the-art methods.

VISION XL is a framework for solving the inverse problem of high-definition video using latent diffusion models. It optimizes the efficiency and time of video processing through pseudo-batch consistent sampling strategy and batch consistent inversion method, supporting multiple scales and high-resolution reconstruction. Key advantages of this technique include support for multi-scale and high-resolution reconstructions, memory and sampling time efficiency, and use of the open source latent diffusion model SDXL. By integrating SDXL, it achieves state-of-the-art video reconstruction on various spatiotemporal inverse problems, including complex frame averaging and various combinations of spatial degradation such as deblurring, super-resolution and inpainting.

Tencent HunyuanVideo is a breakthrough video generation model. It has 13 billion parameters. It is currently the largest Hunyuan video model with the most parameters and the strongest performance among open source models. The model is able to generate videos with strong physical accuracy and consistent footage, provide a hyper-realistic visual experience, and freely switch between real and virtual styles. It has director-level camera operation capabilities, achieving seamless connection between artistic shots and perfectly integrating real effects and virtual scenes. At the same time, HunyuanVideo follows the laws of physics, greatly reducing the sense of dissonance, and through the design of native cutting lenses and continuous movements, users can complete smooth creation with simple instructions, inspiring unlimited creativity and inspiration.

ConsisID is an identity-preserving text-to-video generation model based on frequency decomposition, which generates high-fidelity videos that are consistent with the input text description by using identity control signals in the frequency domain. The model does not require tedious fine-tuning for different cases and is able to maintain the consistency of the identities of the people in the generated videos. The proposal of ConsisID has promoted the development of video generation technology, especially in terms of process without adjustment and frequency-aware identity maintenance control scheme.

Allegro-TI2V is a text-to-video generation model capable of generating video content based on user-provided prompts and images. The model has attracted attention for its open source nature, diverse content creation capabilities, high-quality output, small and efficient model parameters, and support for multiple accuracies and GPU memory optimization. It represents the current cutting-edge progress of artificial intelligence technology in the field of video generation and has important technical value and commercial application potential. The Allegro-TI2V model is provided on the Hugging Face platform and follows the Apache 2.0 open source protocol. Users can download and use it for free.

SoraVids is an archive library of the video generation model Sora based on the Hugging Face platform. It contains 87 videos and 83 corresponding tips that were publicly displayed before OpenAI revoked the API key. These videos are all MIME type video/mp4 with a frame rate of 30 FPS. The background of SoraVids is OpenAI's video generation technology, which allows users to generate video content through text prompts. The importance of this archive is that it preserves videos generated before the API key was revoked, providing a valuable resource for research and education.

PPLLaVA is an efficient large-scale language model for video that combines fine-grained visual cue alignment, user-instructed convolutional-style pooling of visual token compression, and CLIP context extension. The model establishes new state-of-the-art results on datasets such as VideoMME, MVBench, VideoChatGPT Bench, and VideoQA Bench, using only 1024 visual tokens and achieving an 8x increase in throughput.

The NVIDIA AI Blueprint for Video Search and Summarization is a reference workflow based on NVIDIA NIM microservices and generative AI models for building visual AI agents that understand natural language cues and perform visual question answering. These agents can be deployed in a variety of scenarios such as factories, warehouses, retail stores, airports, traffic intersections, etc. to help operations teams make better decisions from the rich insights generated from natural interactions.

MiniMates is a lightweight picture digital human-driven algorithm that can run in real time on an ordinary computer and supports both voice-driven and expression-driven modes. It is 10-100 times faster than liveportrait, EchoMimic, MuseTalk and other algorithms on the market, allowing users to customize their own AI partners with very little resource consumption. The main advantages of this technology include extremely fast experience, personalized customization, and the ability to be embedded in the terminal, eliminating dependence on Python and CUDA. MiniMates follows the MIT license and is suitable for application scenarios that require fast and efficient facial animation and speech synthesis.

Mochi 1 is a cutting-edge open source AI video generator developed by Genmo that allows creators to generate high-quality, photorealistic videos using text and image cues. Mochi 1 makes AI video generation easy for everyone with its superior prompt following capabilities and smooth motion effects. It is designed to compete with other models in the industry, giving creators more control and better visual results.

Allegro is an advanced text-to-video model developed by Rhymes AI that converts simple text prompts into high-quality short video clips. Allegro's open source nature makes it a powerful tool for creators, developers, and researchers in the field of AI video generation. The main advantages of Allegro include open source, diverse content creation, high-quality output, and small and efficient model size. It supports multiple precisions (FP32, BF16, FP16), and in BF16 mode, the GPU memory usage is 9.3 GB and the context length is 79.2k, which is equivalent to 88 frames. Allegro's technology core includes large-scale video data processing, video compression into visual tokens, and extended video diffusion transformers.

sync. is a video lip sync tool that leverages artificial intelligence technology to synchronize lip movements in any video with any audio to achieve natural, accurate and instant lip matching. The tool not only provides content creators, podcasters, and YouTube channel owners with the possibility to repurpose old content, but also helps developers integrate sync. functionality into their applications through its development tools, thus accelerating the global impact of their products. Additionally, sync. supports multiple languages ​​and can easily translate content into any language with life-like, instant lip sync effects.

Pyramid Flow is an efficient video generation modeling technology based on flow matching methods and implemented through autoregressive video generation models. The main advantage of this technology is that it has high training efficiency and can be trained on open source data sets with low GPU hours to generate high-quality video content. The background information of Pyramid Flow includes that it was jointly developed by Peking University, Kuaishou Technology and Beijing University of Posts and Telecommunications, and related papers, codes and models have been published on multiple platforms.

LLaVA-Video is a large-scale multi-modal model (LMMs) focused on video instruction tuning. It solves the problem of obtaining large amounts of high-quality raw data from the network by creating a high-quality synthetic dataset LLaVA-Video-178K. This dataset includes tasks such as detailed video description, open-ended question and answer, and multiple-choice question and answer, and is designed to improve the understanding and reasoning capabilities of video language models. The LLaVA-Video model performs well on multiple video benchmarks, proving its effectiveness on the dataset.

Runway API is a powerful video modeling platform that provides advanced generative video models, allowing users to embed Gen-3 Alpha Turbo into their products in a safe and reliable environment. It supports a wide range of application scenarios, including creative advertising, music videos, film production, etc., and is the first choice of the world's top creative professionals.

Vchitect 2.0 (Dream Building 2.0) is an advanced video generation model developed by Shanghai Artificial Intelligence Laboratory, aiming to give new power to video creation. It supports 20-second video generation, flexible aspect ratios, generation spatial temporal enhancement, and long video evaluation. Through its advanced technology, Vchitect 2.0 is able to convert still images into 5-10 second videos, allowing users to easily transform photos or designs into engaging visual experiences. In addition, Vchitect 2.0 also supports the evaluation of long video generation models. Through the VBench platform, it provides comprehensive and continuously updated evaluation rankings, supporting a variety of long video models, such as Gen-3, Kling, OpenSora, etc.

CogVideoX is an open source video generation model developed by the Tsinghua University team that supports video generation from text descriptions. It provides a variety of video generation models, including entry-level and large-scale models, to meet different quality and cost needs. The model supports multiple precisions, including FP16 and BF16. It is recommended to use the same precision as the model training for inference. The CogVideoX-5B model is particularly suitable for scenarios where high-quality video content needs to be generated, such as film production, game development and advertising creative.

AvatarPose is a method for estimating the 3D pose and shape of multiple closely interacting people from sparse multi-view videos. This technique significantly improves the robustness and accuracy of estimating 3D poses in close interactions by reconstructing each person's personalized implicit neural avatar and using it as a prior to refine the pose through color and contour rendering losses.

ExAvatar is a new 3D full-body dynamic expression model that combines the full-body driving capabilities of SMPL-X and the strong appearance modeling capabilities of 3DGS. It can be created through a simple mobile phone scan and supports animation rendering of various postures and expressions. ExAvatar’s hybrid representation method improves the naturalness of facial expressions, reduces artifacts for new expressions and poses, and makes the model fully compatible with SMPL-X’s facial expression space.

CogVideoX-2B is an open source video generation model developed by the Tsinghua University team. It supports video generation using the English prompt language, has 36GB of inference GPU memory requirements, and can generate 6 seconds long, 8 frames per second, and 720*480 resolution videos. This model uses sinusoidal position embedding and currently does not support quantitative reasoning and multi-card reasoning. It is deployed based on Hugging Face's diffusers library and can generate videos based on text prompts, which has a high degree of creativity and application potential.

Tora is a diffusion transformer (DiT)-based video generation model that enables precise control of video content dynamics by integrating textual, visual and trajectory conditions. Tora's design takes full advantage of DiT's scalability, allowing the generation of high-quality video content at different durations, aspect ratios and resolutions. The model excels in motion fidelity and simulation of physical world movement, opening up new possibilities for video content creation.

metahuman-stream is an open source real-time interactive digital human model project. It uses advanced technology to realize synchronous audio and video dialogue between digital humans and users, and has commercial application potential. The project supports a variety of digital human models, including ernerf, musetalk, wav2lip, etc., and has functions such as voice cloning, digital human speech interruption, and full-body video splicing.