SIMPLE QUESTION! IDEA!

[Tapyon]

Dear Friend:
Lets talk about something i was wondering...
In this case lets say i want to make a video of someone dancing !! ...

--- I understand the program does it LIVE and that's why the RAM , VRAM , GPU makes a crucial matter here...
but... and here me out ...

WHY?
... if we can have the DEPTH, or we can have the POSE estimation, already in a FOLDER, so lets be say... it can be 400 POSE ESTIMATIONs , ..
is there some ""workflow"" , ""model"" , something ? that allowed me to slowly make the pictures 1 by 1 , not FULL POWER, going slow 1 by 1 , i can stop it if i want, eventually i continue the BATCH !... and sooner or later i will have another folder with 400 REAL PHOTOS of the "dancing" ... and then EVENTUALLY i would say another step, that put that into 10 frames chunks, and then eventually those chunks into bigger chunks,.. (and i can transfer this to other examples, but the idea is the same)

I would guess this might help people with less power , and also in other cases, when i don't want my PC completely trying to RUN a workflow like if he is competing to beat the world record!! .. and then telling me that didn't have space enough!
when it can be making 1 frame at the time, i check picture 1 to 10 is doing GOOD, continue the batch, and saving it for a SLOWER , but steady peace, reliable, end work.

Hope what I'm trying to ask is clear, i understand would be but and maybes, that if the video keeping track of the face, the clothes, its only the movement, this and that!
but question still reminds, probably with all the but and maybes still more reliable then making it NOW , FULL THROTTLE
...
wait for some insights and thanks for all.

Tapyón!

[ltdrdata]

That approach is possible, but the issue with generating frames separately is that it breaks the continuity between them, resulting in a final video that looks like a series of disconnected frames.
(Such attempts have been made since the early stages, and in some cases, this method is used for very long videos, compromising quality in the process.)

This is why current video generation techniques aim to generate the entire sequence in one go whenever possible.