Spent two nights messing around with this Uni3C thing, just jotting down some quick notes. It’s a set of tools from Alibaba, plugged into ComfyUI as a control module. The main deal is it extracts camera motion from a reference video and applies it to your own static image or scene—no manual keyframing or rigging needed. I ran it with Wan2.1 I2V 14B 720P fp8 weights, paired with the Uni3C controlnet fp16. Dragged the workflow in and got a sea of red nodes—classic, just hit Manager to install the missing ones.
One heads-up on VRAM: the dev says minimum 24G, but on my 24G card, running 720 resolution with frames maxed out at 125 was already tight. If you can go 48G, do it. The input video needs obvious camera movement—static shots won’t extract anything. For the input image, go for something clear, front-facing, and close-up; too many tiny details and it won’t hold up. The first batch might not be great, so just gacha-style re-roll a few times and pick one. Overall, it’s pretty easy to get started—default settings work fine.