Anonymous
6/24/2025, 3:20:00 PM
No.105689852
>>105664855
I'm trying out WAN 2.1 VACE I2V with self-forcing LoRA and I just can't get it to make a character from reference move in accordance with control video while staying true to reference. I tried extracting both the pose and depth map and combining them into one embedding, but it just warps everything to look kind of like the control video with a reskin. Setting the strength lower makes it ignore the controls. Using just start or start and end frames doesn't help either with some motions - particularly those involving Z axis. It's like it can't portray anything that has to do with things changing relative positions when it comes to depth.
I'm trying out WAN 2.1 VACE I2V with self-forcing LoRA and I just can't get it to make a character from reference move in accordance with control video while staying true to reference. I tried extracting both the pose and depth map and combining them into one embedding, but it just warps everything to look kind of like the control video with a reskin. Setting the strength lower makes it ignore the controls. Using just start or start and end frames doesn't help either with some motions - particularly those involving Z axis. It's like it can't portray anything that has to do with things changing relative positions when it comes to depth.