"He doesn't look like himself.
Want to use AI to make in-depth fake videos? This technology is really inexpensive now. Many people think that this thing will cost a lot of money, but in fact, it can be played for thousands of dollars. For example, using Coze to build a workflow and adding some AI tools, ordinary people can get started.
Let's start with the most critical thing-finding material. You have to have the original video. For example, if you want to imitate a certain star, find a video of his high-definition speech. The more the better, at least tens of minutes. Why? AI needs to learn his mouth shape, expression, and voice. There is too little material, and it will be weird to make it.
Then came the sound. Many people ignore this. The voice was wrong, no matter how similar it was, it almost meant it. You can use the AI voice changing tool to change your voice to the voice of the target person. There are many such tools nowadays, and some can also adjust the speed and tone of speech.
Then there is video synthesis. There are two ways here. One is to directly use AI face-changing tools, such as Deepfake software. The other is to use workflows, such as Coze's script, to sync voices and expressions to new videos. The first type is simple, but easily exposed. The second trouble point is much better.
Let's talk about how to avoid traps. First, don't use too long sentences. When AI changes faces, long sentences are easy to get stuck. Second, the light must be consistent. When you shoot a new video, the background light and face light should be as close as possible to the material. There is too much difference, AI cannot learn. Third, don't exaggerate your actions. Large movements will make the synthesis worse.
Recently, there was news that a certain Internet celebrity used AI to forge an interview video of another Internet celebrity. Everyone believed it at first, but later discovered that it was fake. What does this mean? This shows that this technology is becoming more and more realistic. But realism does not mean perfection. If you look carefully, you can still discover flaws. For example, the eyes are not right, or the fingers are too stiff.
If you want to make this kind of video, it is recommended to start simple. For example, try changing your face to a still picture first. Wait until you are proficient, then make dynamic videos. Don't challenge high difficulties at the beginning and you will easily fail.
There is another key point-legal risks. Many countries have begun to regulate deep counterfeiting technology. For example, using other people's faces to make videos may involve infringement. It is best to check relevant regulations before using.
Talk about tools. There are now ready-made platforms on the market, such as Runway and Synthesia. They encapsulate complex processes, so you just need to upload material, choose a template, and you can generate videos. But such platforms are expensive, with a monthly fee of several hundred yuan. If you're just playing games, you might as well build your own set. Use open source tools and add some scripts to save a lot of costs.
If you want to do a good job in deeply faking videos, patience is more important than technology. Try it a few more times and you will slowly find the trick. Don't be afraid of failure, it won't look good the first time. But as long as you persist, you can always make decent works. "
