[Guide] Vid2Vid SDXL with 4 steps lighting Lora

This workflow is based on @Inner_Reflections_AI's LCM wf, the works of @jboogx.creative and stuff I'm picking up a long the way here at the Banodoco discord and in general

The workflow is very straight forward, has IP-adaptor and only 2 controlnets - Depth and Temporalnet.



I'm still debating over the upscale method and hope to find something different, but that can wait for another day i guess :P

Get the lightning loras here: https://huggingface.co/ByteDance/SDXL-Lightning/tree/main Dynavision model from civit: https://civitai.com/models/122606/dynavision-xl-all-in-one-stylized-3d-sfw-and-nsfw-output-no-refiner-needed
TemporalNet CN here - https://huggingface.co/CiaraRowles/controlnet-temporalnet-sdxl-1.0/tree/main

Important stuff that came out during the testing -
Negative prompt doesn't work well or doesn't work at all in low CFG, so don't count on it to keep your video modest...
Not all SDXL models are created equal and some just don't work with the hotshot animated model.
If you increase steps to 6 (like it is on this WF, they you might not even need the upscaler...)
Happy generations!

I'll be happy to see what you made with this WF, and also, if you like it, you are welcome to have a look at more stuff i'm doing at my Youtube channel https://www.youtube.com/@Just_gili and/or my IG: @gilibenshahar

but really it's all about the creation! so please do something awesome with it!