There are several community doggystyle loras and they all work well so make sure to check them out, but mine is trained specifically on standing position with or without breast grab because this is my fav and I want to produce it consistently. The concept seems easy compared with previous loras I trained, because technically it's just some hips going back and forth, but getting it work without messing up faces or hands were incredibly hard I think probably because of the intense facial expressions in my training dataset.

As always I don't have a trigger word, to create the motion:

thrusts his body/genital/penis back and forth (violently,rhythmically) into her

Hand motion (optional):

fondle/caress/touch breasts with (sweeping touch|encompassing touch) (from bottom)

Example:

Intimate shot of a sex scene. A man behind violently thrusts his body back and forth rhythmically into the girl's body. The male slowly fondles and caresses her breasts with both hands, arousing the female.

You can add facial expressions, but I would not recommend so up front because it can sometimes turn the face into a white girl face due to skewed dataset:

open-mouth expression, arousal, rolling eyes

About T2V & Hunyuan:

• I think T2V models need a lot of static pictures for constructing the first frame, and I don't have a good workflow/process to do it, so temporarily I have decided to give up pursuing training t2v model.

• I got Hunyuan T2V to work, but I can't seem to get i2v to work and I have reached out to some folks for help. If you use Musubi trainer and have successfully made i2v loras please let me know how you were able to do it as I'm having trouble adding CLIP model option. Thanks!