Vace Extension Flow Image or ref to video

This workflow generates videos with vace ref images or vace start frames and is designed to extend the video via batching.  With knowledge you can extend the flow to do more batches.  

This generates 1 initial video, then extends it.  

This can do 2 ref subjects.

Self Forcing models

Any vace 14b model can use the

Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors

from https://huggingface.co/Kijai/WanVideo_comfy/tree/main

full model - 28gb

https://huggingface.co/lightx2v/Wan2.1-T2V-14B-StepDistill-CfgDistill/tree/main

1.3b self forcing (fast and decent)

https://huggingface.co/lym00/Wan2.1_T2V_1.3B_SelfForcing_VACE/tree/main

1.3b Lora works with any vace 1.3b model and the model itself works alone.

I use gguf q8 14b with the lora and

I use 1.3b vace with lora or the full self forcing model.

Notes in the flow have models and info needed.

Enjoy :)

Issues with

Update comfy ui and its frontend.  This was just fixed recently.  

This was/is due to the set/get nodes having issues with some front ends of comfy ui.

Update to latest front end for comfy ui should fix it.

Newest front end that works with set/get

--front-end-version Comfy-Org/[email protected]

older front end that worked - all between these two dont.

--front-end-version Comfy-Org/[email protected]