Russia
Joined: Aug 13, 2022
Post Count: 568
Status:
Offline
thoughts on AI for video
I'm sure it's a mistake to consider AI your friend. He will never become obedient, he always plays his game, you always have to be careful with him and keep him at a distance. It's an animal in fluffy colorful packaging that will run away or bite at the first opportunity. This wolf will never become tame and trained. That's the main thing. Therefore, AI games need to be done carefully and think through every step. I decided to make a big video about the interior of the apartment in which I am currently building. This video used only static renderers (6 pieces) made using SH3D and two real photos from the construction site. I tried to use AI to simulate the movement of a video camera, so far I have been able to do this using HAILUOAI, there is a unique tool for adjusting the movement of the video camera relative to the photo. It is very important to understand that this is precisely the movement relative to the image, and not the movement inside the space that is depicted in the photo. I often confused this and got the wrong results. Therefore, consider how the camera moves over the photo, especially if it is a difficult angle. Always trying to give an AI a task ended in a surprise, the AI never did everything as I wanted. Never. Therefore, cooperation with AI is like a dolphin show. It seems to the audience that the dolphins follow the trainer's commands, but in fact the dolphins just enjoy playing their games according to their nature, and the trainers adjust to the course of the show. I read this interesting interview with dolphin trainers, it's exactly like that)). Of course, when working with AI, you need to have a work plan, but be prepared for it to change, which is normal. The main thing is to get an impressive result, and this result may be unexpected. I used two more neural networks in my work. I skipped every piece of the received video from AI through removing the watermark and improving the video quality. VMAKE.AI и X-DESIGN.COM . Yes, sometimes AI really shows the volume very truthfully and truthfully makes the camera move in the interior, but only at the very beginning. Then, anyway, the AI puts on its own show and makes surprises. Therefore, it is often possible to use only half of this video movement, because the other half is completely untrue. And now an honest, obedient tool comes to the rescue - a video editor. Everything is fair there. This is not wild AI. The desired cropped piece can be stretched over time, add fades, and cross-overlap. It's a more comfortable job here, and the man is the boss. And in addition to AI work, you can add video effects from the video editor itself (the beginning from another section of the video or photo, the final frame in another place of the video or photo. This is another kind of movement that can be combined in one movie. In my video, all the frames are made with these types of movements, which makes it almost no different from stationary renderers and AI videos. Please watch this video, it can be a good alternative to the SH3D video. All efforts are applied to the renderers, the video was made using AI and a video editor. This is a great time saver. It took me three days to do all the work, because HAILUO's free quota is only three attempts a day, I had to wait three days and make almost no mistakes. Of course, all this can be done using SH3D, but videos with such angles and lighting can take many weeks of computer time. I tried, the result is very realistic, but the computer heats up like a blast furnace. Therefore, it is better to put it in the clutches of AI. Carefully and without trust.
Russia
Joined: Aug 13, 2022
Post Count: 568
Status:
Offline
Re: thoughts on AI for video
---- I'm trying to puzzle the AI with new tasks. Now there are live pictures on the laptop screen (at the end of the video). You need to work with AI like a hot horse. You can't whip him and give him a complicated text at once. Otherwise, he's off at a gallop. You need to give him a task carefully, patting him with the palm of your hand. If you add, for example, flying flocks of birds outside the window to the task about the laptop screen, the room will turn into chaos with a flying laptop. This is the first time I showed this video to a real client (these are two real rooms with custom furniture options). AI helped me with my work in this case. This method can be continued.
Russia
Joined: Aug 13, 2022
Post Count: 568
Status:
Offline
Re: thoughts on AI for video
Dear Wendell_Burke shared his space station project in his topic . It is very difficult to work with many models. I tried to make renderers, and this large project turned out to be too large for the capabilities of SH3D. But I tried to connect AI and the result was this video. Here are two static renderers from two points. Several short videos processed with the help of AI (I removed the background and applied layers in the video editor). AI was able to make a good video for free. A static picture turns into movement with the help of AI HAILUO.
Russia
Joined: Aug 13, 2022
Post Count: 568
Status:
Offline
Re: thoughts on AI for video
I tried to assign a person with a smartphone that will be reflected in the glass of the cabinet. And he made it more difficult for the AI to change the image on the TV. In this short video, two neural networks were used - PixVerse end Hailuo. And a zoom for the static render SH3D at the beginning. Two neural networks increase the number of free attempts. Both short pieces of video were processed through watermark removal and quality improvement.
Russia
Joined: Aug 13, 2022
Post Count: 568
Status:
Offline
Re: thoughts on AI for video
the renderers are made in SH3D. Video of three rooms using AI.
The eye was animated with the help of AI, part of the furniture was made with the help of AI. The dog is real, the owners actually have the same dog. Made with the help of AI.
