Scrubbed Wan Vace workflow - v1.0 Showcase

This is associated with some projects I make, and I'm often asked for the workflow. The workflow notes that I include in those posts are probably more useful than the actual .json file:

Nodes. The most unconventional node that I use is the "Image Batcher by Index Pro V2"; it lets you insert keyframes in WanVace using images or video, and lets you determine if they are masked or not during the generation process. A lot of my process pivots around this ability.

Loops. A. Create the base animation. B. Create a bridge animation with similar prompts and settings as the base, but putting the last 16 frames from the base at the beginning and the first 16 frames from the base at the end (if not 16 frames, make sure its enough frames to give continuity of motion). C. In software like Openshot, load the base and bridge animations. D. Split the bridge animation down the middle into two clips. E. Take the first half of the bridge and line it up with the end of the base; take the last half of the bridge and line it up with the start of the base. F. Where the animations overlap, fade from one to the other; this'll mute VAE decoding artifacts and soften any color shifts. When the animation gets to the end of the frankensteined clip, it'll be just before the middle of the bridge animation, and when it starts it'll be just after the middle of the bridge animaiton. Voila- a loop with potentially seamless movement.

My Standard Process. A. Create the first generation using a high quality starting frame, with CFG and loras pumped all the way up (this'll shift the style to the point that the results will be unusuable in any direct sense, but it will get the most out of the movement in the loras). B. Create the second generation by using the same start frame and a blurred depthmap of the video from the first generation, with the CFG and loras knocked down; most of the movement will be captured by the depthmap, and the style won't shift as much. C. If it works as is- cool! If not, identify frames from the result; save the frames as images. D. Do any manual touchups desired on the frame. E. Repeat B, but with the new keyframes added to pull the animation towards the quality you want.

ComfyUI Workflow. I don't update until I'm inbetween projects, so I'm almost always far enough behind that my ComfyUI workflows break in up-to-date installations. The only really novel thing in it is the "Image Batcher by Index Pro V2". Add that to any wan 2.1 vace workflow and it'll do 90% of what my comfyUI workflow does with a lot less headache than trying to decipher the rest of my mess.

Other Tools. In addition to ComfyUI, I use Clip Studio Pro for manual edits, Moho for manual animation, Openshot for compositing, and Topaz for upscaling. I'm using a second-hand 3090.