Why did American Labs seemingly give up on video generation?
Summary
作者质疑美国实验室为何似乎放弃了视频生成模型的竞赛，认为 Google 对 Veo 的投入不足，而目前唯一持续进步的大型视频生成模型似乎只有 Seedance。
Similar Articles
AI video generation models still have a long way to go
The author criticizes AI video generation models like Seedance 2.5 for weak prompt understanding and errors such as misspelling, suggesting these models still face significant challenges compared to LLMs.
@_TobiasLee: Seed 2.1 from Bytedance achieved impressive results on two of our benchmarks. Claw-Eval (Multimodal, https://claw-eval.…
ByteDance's Seed 2.1 model achieved strong results on multimodal agentic (Claw-Eval) and long video understanding (Video-MME) benchmarks, though a gap remains between perception and agentic capabilities.
ByteDance's New AI Video Model Can Make 30-Second Clips From a Single Prompt (2 minute read)
ByteDance unveiled Seedance 2.5, an AI video model that can generate 30-second 4K clips from a single prompt or up to 50 reference pieces, rivaling OpenAI and Google's offerings.
Open-Sora 2.0: Training a Commercial-Level Video Generation Model in $200k
Open-Sora 2.0 is an open-source video generation model trained for only $200k, achieving commercial-level performance comparable to leading models like HunyuanVideo and Runway Gen-3 Alpha, with the aim of democratizing advanced video creation.
Fuel your creativity with new generative media models and tools
Google announces Veo 3 and Imagen 4, next-generation video and image generation models with significant capability improvements including audio generation and enhanced physics simulation. The company also introduces Flow, an AI filmmaking tool, and expands access to Lyria 2 for music creation.