workflow that either :
lets you draw a mask to video based off the first frame
or detects a watermark and creates a mask with YOLOv8 bounding box for each frame
V2 uses SD 1.5 + VACE + WAN to inpaint over it
V1 uses Flux + VACE + WAN to inpaint over it
do a run of only step 1 (use option 1 or option 2) to get the masks
then toggle to the step 1 option you selected in the intermediary step
then enable steps 1-2 and get a good inpaining
then enable with steps 1,2 and 3 and run
used yolov8 model
https://civarchive.com/models/311869?modelVersionId=350003
accumulating for loop node
https://github.com/BadCafeCode/execution-inversion-demo-comfyui
Description
v2
-has args for 14b version and notes the blocks VACE switch with, its slower but better
-I also switched from Flux to good ol' SD.15 as its way faster, uncensored, and the part it is used does not need to be perfect. WAN refines it anyway
-I added a section to trim and or resize the video as some of the videos I was testing were too big to process