Changes in version 3.7 (version 2 is the simple version)
This workflow builds on the prior version to address various drawbacks:
The final frame of each video gets progressively rougher
LoRAs were set at the start and used for the whole flow
Faces change when characters go in and out of frame
You need to load an initial keyframe
In this version we add options to:
Use Flux to refine keyframes, improving the quality (GGUF or Safetensors switch)
Inject LoRAs at whichever step (so you can swap loras in each shot)
Use ReActor to sync faces back to the ones in the first keyframe
Use Flux to generate an initial keyframe (or load an image)
All of these are switchable and the main flow will use any of the options that are switched on.
You must set all the prompts you're going to use manually (for video and keyframe generation and refinement).
You can switch off face swap for post processing (it will look bad if any part of the face gets obscured and it takes a long time)
