It looks very practical. Can you provide a version and method supported by comfyui?
#1
by RedHn - opened
THX!!!!
The same idea on a much smaller model that is also multimodal. LightX2V's second adapter is trained on Qwen3-VL-8B-Instruct, so where the 27B has to be told what a reference frame contains, this one is shown the frame and writes the alignment line from what it sees. It covers four tasks rather than one, and fits on a card the 27B cannot go near.