V11 - Updated for compatibility with recent ComfyUI versions (ComfyUI Frontend versions > 1.40.0.x totally borked subgraphs).
V10 - Use with ComfyUI Frontend versions <= 1.40.0.x
This is my take on an "ultimate WAN 2.2 I2V / SVI" workflow.
I can only recommend using this WF if you are also using ComfyUI-Easy-Install. It seems that the some of the very popular "must-have" nodes I've used, which are included in this, may be problematic for users who are NOT using it.
Convenience features:
- Global LoRAs option. Each gen step can have additional LoRAs.
- Global prompt prefix/suffix option. Each gen step can have this customized.
- Each gen step has option to "Save latent" (can be loaded later for SVI extension)
- Color correction option for each gen step
Can independently run I2V or SVI or both
- Starting point for I2V can be First Frame ***or*** FLF2V (first/last) ***or*** FMLF2V (first/middle/last)
- Start point for SVI can be I2V ***or*** load a saved latent ***or*** load a video file
SVI blocks:
- Can have overrides for Shift, Total Steps, Split Step
- Can be duplicated
My custom Image Resize node can be found here:
https://github.com/altoiddealer/comfyui_essential-er
The "Save/Load" Latents node can be found here:
https://github.com/A1-multiply/ComfyUI-LatentSaver
I also used a few nodes that I like from this node pack:
https://github.com/jamesWalker55/comfyui-various
Recommended models:
SmoothMix V2 - High Noise
DaSiWa Lightspeed SynthSeduction v9 - Low Noise
Recommended parameters:
Shift: 5.0
High noise steps: 3
Low noise steps: 4
cfg: 1.0
Sampler: lcm
Scheduler: beta
Description
FAQ
Comments (3)
So I get to be the guinea pig for this one and lose 2 weeks of my life fixing three hundred unmatching dependencies ?
TBH - the only reason I uploaded this is because someone asked me to share my workflow. Now you are certainly overexaggerating about the three hundred unmatching dependencies, this workflow uses only extremely popular node packs with only a few exceptions which I have direct links to in comment blocks. One is my fork of an Image Resize node, which I've isolated to just 1 node in the pack. The other is the custom Load/Save Latent nodes - if you don't want to install those nodes to use that feature, just delete it. The native Comfy Core latent save / load nodes are basically worthless. As for the workflow itself, I've adhered to good practices in trying to keep as much exposed in the main graph as possible. I could expose CFG from the subgraphs, and sampler/scheduler if I figure out a clean way, otherwise you should never have to click into any subgraph.
Feel free to let me know if there's something I'm missing.
I reviewed and I realize that I also used a few nodes from this pack which is not quite as popular, and which I did not mention (will add to description now): https://github.com/jamesWalker55/comfyui-various