Why the rendered Videosequences are always end in the same Starting-Position as on the Control-Image?
Suggestion: A Person with her/his hands on her lap, should wave with her hands.
It doesn´t end, withe waving, but with her/his hands on the lap.
So, many frames are lost for turning back to the starting position.
WHY?
Maybe you put end frame(s) in your workflow? The generated video will aim to end to this frame(s) every time.
this is just a 'getting around' suggestion, but you can pick some frame from the middle of your generated video where the hands are actualy in the air waving and use that frame as end_image ;-)
Maybe you put end frame(s) in your workflow? The generated video will aim to end to this frame(s) every time.
No, I don't...
this is just a 'getting around' suggestion, but you can pick some frame from the middle of your generated video where the hands are actualy in the air waving and use that frame as end_image ;-)
If I render a complete video, just for taking a frame out of the Middle: Isn't that contra productive?
this is just a 'getting around' suggestion, but you can pick some frame from the middle of your generated video where the hands are actualy in the air waving and use that frame as end_image ;-)
If I render a complete video, just for taking a frame out of the Middle: Isn't that contra productive?
nope, just bad explanation from my side - > I meant - 1st generation - you make the video you are not satisfied with (aka end with hands in lap) - and from that video you extract that frame, so you can use it in 2nd generation, now with the end frame. So you will get your desired result at the end... but at the end of 2nd generation, I know... Do you understand me better now?
It's like the model takes the frames in a sine wave: from start to end, back to start.
I.E.: WORKFLOW: THE WOMAN SIT HERSELF DOWN. FACT: THE WOMAN SITS DOWN, AND STANDS UP AGAIN. I just don't understand the reason of that behaviour...
There's also a lot of loss of frames by that.🤷♂️
A problem could be that your prompts are too short. Everything you describe in the prompt is maybe archived within the first half of the video so the model has no further informations on what to do next and goes back torwards the starting position.
Try something like: THE WOMAN SIT HERSELF DOWN and does not move anymore.
I came across this with my video extend workflow where like for example you prompt something like "He spills a glass of milk on the table." - and toward the end the milk would magically evaporate. After some poking around, I tracked it down to something funky with the Wan Video VACE Start To End Frame node. I suspect internally it might be doing something like defaulting to the start image to the end image if no ending image isn't provided. I was able to get around it by copying this bit out of the default VACE workflow in Comfy:
