Vace1.3B model struggles with characters, but its power is undeniable.
Let's prepare for the arrival of the mighty Vace14B!
The principle is to extract frames with a certain value from the previous video to guide the next video,If possible, you can keep copying the workflow to extend the video duration
You can extract more frames or fill in different prompt words for each workflow segment to achieve your goal
In theory, you can generate a movie ：）
Description
FAQ
Comments (9)
Great scheme! Really creates long videos, more than 10 seconds. Thank you!
Is it possible to finalize a little scheme so that it is not necessary to cut the picture? When trying to set "Keep Ratio", an error appears ...
RuntimeError: shape '[65, 160, 8, 76, 8]' is invalid for input of size 50918400
Please check that your input dimensions are multiples of 16 and not too large.
Thank you, really depends on the size of the picture.
tutorial. some of us are novices :)
There are links in the workflow and introduction
how much VRAM it cost? Does it work well with 16gb vram?
Yes
To begin with, I am using the latest nightly ComfyUI as of the date of this post. It is not 100% clear what this workflow does. It has quite a few nodes. Why so many text input boxes? Can you prompt at each level? Do you have to enable the different text input nodes?? I thought the two load images were possibly start/end frames, but that is not what happens, so what does the second load image do? Is the 2nd image disconnected by default, and you need to connect it if you want to use the end image? Are you able to use the different modes of ACE, or is this strictly a text-to-video workflow with start and end frames and video extending? Top to bottom, what does each level do? Are you just extending the video at each level by the same number of frames as selected at the top, and finally upscaling and joining at the end? ComfyUI dumps on the Scale to Megapixels node after about 25 percent (I was using 81 frames as I thought that was going to be the total number of frames and not the total number at each level. It does not just crash Comfy; the whole RunPod crashes. I can't just restart Comfy, I have to terminate the pod and restart. Are there recommended upscale models? Maybe add a switch to be able to toggle on/off the different levels and whether or not to use the end image? An explanation of everything that is going on would be helpful.
Your problem can basically be solved in the V2.0 version I released