经济观察网 9月25日，生数科技正式发布新一代图生视频大模型Vidu Q2，在表情变化、推拉运镜、生 成速度及语义理解方面有所进展， 主要包括图生视频、首尾帧视频、时长可选（2-8s）、电影大片及闪 电出片两种模式。 ...

Core Viewpoint - The article evaluates the capabilities of nine representative image-to-video AI models, highlighting their advancements and persistent challenges in semantic understanding and logical coherence in video generation [2][7][50]. Group 1: Evaluation of AI Models - Nine models were tested, including Google Veo3, Kuaishou Kling 2.1, and Baidu Steam Engine 2.0, covering both newly launched and mature products [7][8]. - The evaluation focused on real-world creative scenarios, assessing models on criteria such as image quality, action organization, style continuity, and overall usability [9][14]. - The testing period was in August 2025, with a standardized prompt and conditions for all models to ensure comparability [13][9]. Group 2: User Perspectives - Young users, who are not professional video creators, expressed a need for easy-to-use tools that can assist in daily content creation [3][4]. - The evaluation was conducted from a practical and aesthetic perspective, reflecting a generally positive attitude towards AI products [5]. Group 3: Performance Metrics - The models were assessed based on three main criteria: semantic adherence, physical realism, and visual expressiveness [14][21]. - Results showed that Veo3 and Hailuo performed best in terms of structural integrity and visual quality, while other models struggled with semantic accuracy and physical logic [17][21]. Group 4: Specific Use Cases - The models were tested across various scenarios, including workplace branding, light creative expression, and conceptual demonstrations [11][16]. - In the workplace scenario, models were tasked with generating videos for corporate events, while in creative contexts, they were evaluated on their ability to produce engaging and entertaining content [11][16]. Group 5: Limitations and Future Directions - The evaluation revealed significant limitations in the models, particularly in generating coherent narrative sequences and adhering to physical laws in complex scenes [39][50]. - Future developments are expected to focus on enhancing the models' ability to create logically complete segments, integrate into creative workflows, and facilitate collaborative storytelling [53][54][55].

Core Viewpoint - The rapid iteration of AI video models has created a landscape where users can easily generate videos, but practical application remains a challenge for ordinary users [2][3][4]. Group 1: User Needs and Model Evaluation - Many users require clear narratives, reasonable actions, and smooth visuals rather than complex effects [4][6]. - The evaluation focuses on whether these models can solve real problems in practical applications, particularly for novice content creators [5][7]. - A series of assessments were designed to test the models' capabilities in real-world scenarios, emphasizing practical video content creation [8][9]. Group 2: Model Selection and Testing - Eight popular video generation models were selected for testing, including Veo3, Hailuo02, and Jimeng3.0, which represent the core capabilities in the current video generation landscape [11]. - The testing period was set for July 2025, with specific attention to the models' performance in generating videos from text prompts [11]. Group 3: Evaluation Criteria - Five core evaluation dimensions were established: semantic adherence, physical laws, action amplitude, camera language, and overall expressiveness [20][25]. - The models were assessed on their ability to understand prompts, maintain physical logic, and produce coherent and stable video outputs [21][22][23][24][25]. Group 4: Practical Application and Limitations - The models can generate usable visual materials but are not yet capable of producing fully deliverable commercial videos [57]. - Current models are better suited for creative sketch generation and visual exploration rather than high-precision commercial content [65]. Group 5: Future Directions - Future improvements may focus on enhancing structural integrity, semantic understanding, and detail stability in video generation [60][61][62]. - The rise of image-to-video models may provide a more practical solution for commercial applications, bypassing some of the challenges faced by text-to-video models [62].