Today's featured cutting-edge AI information, welcome to read 👇
💡 Alibaba releases QwQ inference model, with multiple capabilities comparable to OpenAI's o1, showing better performance in mathematical understanding and problem-solving, but still has limitations in language switching and other aspects.
🎥 ConsisID model achieves character-consistent video generation from a single portrait photo, currently with average generation quality but good character consistency.
🎬 Google launches Generative Omnimatte video decomposition technology, capable of splitting videos into RGBA layers with transparency, enabling background modification, character replacement, and other functions.
Latest News
1. Alibaba launches QwQ inference model based on QWen.
Multiple capabilities are comparable to OpenAI's o1 model, and even surpass o1 in mathematical problem-solving and understanding tests.
Detailed introduction: https://qwenlm.github.io/zh/blog/qwq-32b-preview/
Try it online: https://huggingface.co/spaces/Qwen/QwQ-32B-preview
Currently in preview version, with some limitations such as language switching, reasoning loops, security, and capability differences.
Cutting-edge Technology
1. A video generation model maintaining character consistency: ConsisID.
Only requires one reference portrait photo to generate high-quality videos through text while maintaining character consistency.
Detailed introduction: https://pku-yuangroup.github.io/ConsisID/
Try it online: https://huggingface.co/spaces/BestWishYsh/ConsisID-preview-Space
After testing, the generated video quality is average, but character consistency is maintained quite well, worth paying attention to.
2. Google releases a video decomposition technology: Generative Omnimatte.
Can decompose input videos into a series of RGBA layers with transparency, each layer containing a complete object, which can be either objects or characters.
Detailed introduction: https://gen-omnimatte.github.io/
This technology can be applied to many scenarios, such as modifying backgrounds, replacing characters, removing specific areas, etc.
