for classifier free guidance would it be better to use a blank condition image for negatives?

[jaymefosa]

        latents_pose = poseguider(pose_condition)
        # latents_pose = rearrange(latents_pose, "(b f) c h w -> b c f h w", f=video_length)
        if do_classifier_free_guidance: latents_pose = latents_pose.repeat(2,1,1,1) # b c h w

here instead of repeating, would passing zeros through the poseguider and then catting be more appropriate?

[guoqincode]

I haven't tried this, have you?