The kind where AI watches videos with you and chats while watching | ChatVideo is playable online
Now, AI has taken the matter of "watching videos" to a new level.
You can chat with it while watching the video!
For example, my aunt's:
For the basic information of pandas in the video, AI can answer fluently.
Even such intentional and emotional questions as "why the panda raised its hand" and "how does the panda feel" can be done by AI.
Then follow this video:
For the gender of the characters and the sports they do, AI can also easily handle them.
This is the latest project from the general vision platform OpenGVLab (co-created by Shanghai Artificial Intelligence Laboratory and Shangtang Technology) - Ask-Anything .
It has to be said that GPT-4 has not yet been opened for viewing pictures, and the open source community has already set up video chatting.
It's no wonder that netizens called out:
It can no longer keep up with the speed of AI development.
Currently Ask-Anything provides a demo that can be played online, and the operation is also very simple.
Just "drag" or "upload" the video to the specified location in the upper left corner of the page:
For example, if we upload a video of a woman doing yoga, AI will instantly give a theme: yoga .
Then enter the OpenAI API, click "Watch it", let the AI watch the video first; then click "Let's Chat" to start the conversation.
For example, ask "what is she doing", and the AI will say:
Sorry for not giving a clear answer earlier. Since I don't know which scene is in the video, I'll just give a rough description: She appears to be doing a yoga pose . If you have further questions or need more help, please don't hesitate to let me know!
Then ask "when was the video taken":
Based on my analysis and observation of the scene, I am unable to determine the exact date and time this video was shot. What I can tell you, however, is that if it's shot in a modern city, it's most likely a photographic work done to represent the cityscape. But perhaps its time and location could be better inferred if more background information and textual data were available. Hope this helps you get a better understanding.
From the results point of view, it is still relatively intelligent.
As for the technology behind it, according to the official introduction, Ask-Anything currently supports ChatGPT, miniGPT4 and StableLM.
Behind the team
The project is created by OpenGCLab, a general vision open source platform.
This platform was released and open sourced in February 2022 by Shanghai Artificial Intelligence Lab and SenseTime.
It is understood that the OpenGVLab platform is built based on the "Scholar" technology system. The AI supported by it can easily see complex pictures such as playing mahjong, racing motorcycles, and pandas, and the confidence level is greatly improved.
Finally, the demo address is below, interested friends can get started~
Experience address: http://126.96.36.199:7860/