Save the gifs and one piece of good news. Based on the gifs, the conclusions are as follows. 1024x1024 - not available. 883x883 - only two frames per duration: the most popular product is 16 frames. Resolution 308 x 308. 64 frames. The picture starts to shake and the frame itself starts to shake, even though the message says camera is static. 128 frames, everything starts to break down. 256 frames. Everything eventually breaks down. An interesting nuance is that the documentation for the API includes only 64 frames. But in the ChatGPT interface you can do both 128 and 256. But the result is still rubbish. You can attach a reference. This, by the way, dramatically reduces the amount of noise. I put my face in, and there, in high resolution, are the remains of Sora. One more nuance: you can only edit GIFs created from which it has all the metadata. Now the good news. When I pressed my muzzle, I asked him: how is it now with a muzzle? And that's what he answered me. What is happening with politics based on faces and real people? The most important thing here is that there is no longer a complete ban on the faces of real people. But that doesn't mean that "everything is possible". Logic is generally allowed OpenAI's terms of service specifically state that visual capabilities accept images and videos as input, but cannot be used to identify a person, extract private/sensitive information, and cannot reproduce someone's appearance without express consent and the necessary rights. What about public figures? This is exactly where politics has become softer than before. OpenAI's original image generation system map says that the model can generate images of public figures, and OpenAI doesn't completely block it, as older DALL E models did. Rather than an outright ban, more targeted protective restrictions apply.

Save the gifs and one piece of good news. Based on the gifs, the conclusions are as follows. 1024x1024 - not available. 883x883 - only two frames per duration: the most popular product is 16 frames. Resolution 308 x 308. 64 frames. The picture starts to shake and the frame itself starts to shake, even though the message says camera is static. 128 frames, everything starts to break down. 256 frames. Everything eventually breaks down. An interesting nuance is that the documentation for the API includes only 64 frames. But in the ChatGPT interface you can do both 128 and 256. But the result is still rubbish. You can attach a reference. This, by the way, dramatically reduces the amount of noise. I put my face in, and there, in high resolution, are the remains of Sora. One more nuance: you can only edit GIFs created from which it has all the metadata. Now the good news. When I pressed my muzzle, I asked him: how is it now with a muzzle? And that's what he answered me. What is happening with politics based on faces and real people? The most important thing here is that there is no longer a complete ban on the faces of real people. But that doesn't mean that "everything is possible". Logic is generally allowed OpenAI's terms of service specifically state that visual capabilities accept images and videos as input, but cannot be used to identify a person, extract private/sensitive information, and cannot reproduce someone's appearance without express consent and the necessary rights. What about public figures? This is exactly where politics has become softer than before. OpenAI's original image generation system map says that the model can generate images of public figures, and OpenAI doesn't completely block it, as older DALL E models did. Rather than an outright ban, more targeted protective restrictions apply.