well, even the examples barely move, so...

@sunburnt_toasty Right, like, if you're looking for deep crazy movements, you'd be better off training your own lora and spending the required money to do so, lol. I mean, this lora actually looks pretty good, no complaints here, and some of the images down below show good movement, so I know it can be done, but they're paired with other LORA which would probably help. Point is, I appreciate this guy's comment but they could tell from the demo reel that the motion isn't huge, so why mention that they can't get the big strokes? lol. idk man.

If you want deep thrusting I'm your man, JK xD

As said in another thread, in my experience it works best when paired with other loras that already have this motion but do not make the proper deformations otherwise.

use my workflow. just download my latest video

@Bonticarius Can you please provide a link as I cant' find it and need a WF for this. Thanks!

@thomas852002887 just download the video https://civitai.com/images/100091170. also if youre generating in 720p will have an effect and you have to play around with the CFG and add a few loras

Late but you're apparently not supposed to use it vanilla. It will make the proper motions when added with other humping loras but transforms it into anal

If you haven't figured it out by now, there are three tabs above the sample images.

@Walter_B yeah, thanks i did, than your the response tho <3

Do you need an nude input image maybe? Because i just take "normal" pictures and maybe thats exactly the issue because it gots to generate the missing body parts first.

@teeay25225 I can do that with great success. It's all in how you prompt and the weights you use. CFG does make a huge difference as well. Try try using a CFG of 1.2.

For your high and low samplers

Euler/Euler A
Simple/Normal/Beta
High Sampler Settings:
Total Steps: 8
Start on: 0
End on: 3

CFG 1.2

Low Sampler Settings:
Total Steps: 8
Start on: 3
End on : 8
CFG 1 or 1.2

You'll want to incorporate DR3AMALY V2 a lot of the time as it does help a lot even with concepts it doesn't "support." Use a weight between 0.35 - 0.75 for both high and low. If you're using more than 2 lora's per sampler then you most likely want to keep the weights low on all loras with the exception of maybe 1 (usually DR3AMALY).

Daiwa's checkpoints already have A LOT of concepts already built in and can do most things very well, but no checkpoint can do it all. So it just needs a little push.

K3NK's animation helpers and LoRA's are also incredible. Don't be afraid to use LoRA's that seemingly have no relation to the concept you want. That missionary LoRA most like will work very well for PoV doggystyle sex, hell it might EXCEL at it more than a LoRA designed for doggystyle sex. Some trainers are very good and they capture body parts, motions, and other concepts during training that simply apply to other things too.

The worst mistake you can make though is trying to "FORCE" something to happen. If you keep adding more and more tokens then things will get muddled. WAN can take long prose just fine and doesn't very well with it - but if you start getting poetic and use metaphysical phrases like "he thrusted into her abyssal hole like warlocks seeking answers to a riddle using their crooked staves of penetration" then it'll just go fucking nuts and either render a warlock doing that or simple animated the girl smiling like an idiot into the camera.

However you can have supporting phrases: "The man fucks her ass using long intimate strokes that slowly move back and forth inside the woman's asshole; gently exploring her asshole with his penis, feeling her anal tract squeezing his shaft, the friction is visible as it tugs on the skin of his penis."

I've used that prompt to get it to animate the tugging of the flesh surrounding the penis to make it more realistic.

If you want to remove clothing you absolutely need to be clear in the prompt with what you want to happen. Simply saying, "he pulls her pants down" will mostly end up in failure with some good successes.

If you say, "The man enters from the right and grips the girl's pajama pants/bottoms; he tugs on them, pulling the pajama pants down around the girl's knees. The girl's ass bounces as her pajama pants are pulls down, her firm round ass is now exposed. We can see her dark anus as she turns her ass towards the camera."

Something like that will usually render much better.

The starting image does make a difference, no doubt it'd be easier to start with a penis already inserted, but it's very possible to have the model render a penis, have that penis penetrate whatever you want, and then fuck it silly.

And here's a tip that will save your mind:

Once you get a good render going, then save a copy of that workflow and name it "WAN2.2 - ANAL SEX - SIDE/FRONT" or whatever. That way you have a functional base.

You will thank yourself for doing that. Or save the metadata to your renders, you can always load a good render to see what you did.

@DaddyWolfgang Wow thanks for your time and this long answer! The DR34ML4Y lora works very well now since i use a normal checkpoint with speed lora. The Boundbite10 checkpoint caused way too many issues with this lora. You gave some really good advice that DR34ML4Y can help with this anal lora. I render locally since 1 week with a RTX 5080 and this whole thing is indeed very complex and you must invest alot of time to become very good. And some advice here and there can definitely speed things up and save you a ton of bad render.. So thanks again!