Can i inference 1 frames by 150 times to get a full length render results?

[zzz5y]

Hello! I wonder if i can inference 1 frames by 150 times to get a full length render results?
My idea is inference 150s 1frames?But will the results still keep the time-space ?
Sorry for my stupid quesion because i'm new to diffusion!
Thank you sir!

[flymin]

No, you cannot. You may confuse the time axis in a video with the timestep in the diffusion process. They are different.

There are some image-to-video works that modify the image generation models for video generation without training. However, in the formulation of video diffusion models, the video is treated as a standalone variable.

[zzz5y]

No, you cannot. You may confuse the time axis in a video with the timestep in the diffusion process. They are different.

There are some image-to-video works that modify the image generation models for video generation without training. However, in the formulation of video diffusion models, the video is treated as a standalone variable.

Thank you for your quick reply.I got the first paragraph. But the second paragraph still a little confuse to me.

[zzz5y]

No, you cannot. You may confuse the time axis in a video with the timestep in the diffusion process. They are different.

There are some image-to-video works that modify the image generation models for video generation without training. However, in the formulation of video diffusion models, the video is treated as a standalone variable.

Thank you sir ! I think i understand paragraph 2 which means there is no connection between different 2 generated videos?