MISSIONLG - Mission Impossible
Missionary on MiniMax H3, with a real grind. Bodies lined up and pressed together, his hips rolling into her, and she moves with him.
Base H3 can already do missionary. MISSIONLG gives it more active hip motion and keeps the two bodies lined up and pressed together, from the side, from beside her head, from behind him, and from above.
Built 100% from my own hand-checked footage: 272 clips, 17+ source videos, many different couples. Every clip was looked at by eye before it went in.
Quick start
Strength 1.0.
Trigger: MISSIONLG (optional, it works without it).
Base: trained on stock MiniMax H3 (
minimax_h3_fl2va_pruned_int8_convrot). Also tested on Eros Max.Start image: the two of them already together, him on top between her legs. It will pull an upright start down into a grind on its own.
The prompt line to copy
This is the line that gave the best motion in testing. Copy it as is:
She lies on her back with him on top between her legs. He grinds his hips into her in fast, deep rolling circles, pressing his weight down into her, his hips pushing forward and back with each roll.
Want her moving back? Add:
and she lifts her hips to meet each roll.
Camera angles it was trained on
Your start image sets the angle. These match the training:
Filmed from the side of the bed.
Filmed from beside her head, looking down along their bodies.
Filmed from the foot of the bed, behind him.
Filmed from above, looking down at the bed.
From behind him is a standout.
Tips
Speed words: "fast, deep rolling circles" works. "Fast, long and deep" breaks the motion. "Slow" makes it crawl.
Stacking: a motion booster adds energy. Start it low. Keep the whole stack under about 2.0 or limbs start to melt.
Breasts: in this position his chest is usually on hers, so they barely move on their own. Stack your favourite breast physics LoRA for bounce.
Known limits
It holds the camera still. The training footage was all fixed cameras, so it tends to ignore tracking or orbit moves.
Breast redraw: if his arm hides her breast and then moves away, the redraw can come out wrong. A start image with her breasts already visible avoids it.
Description
v1.0
First release.
Trained on stock MiniMax H3, image-to-video.
272 hand-checked clips from 17+ videos, many couples.
Four camera angles: side, beside her head, behind him, above.
Grind motion built in.
Strength 1.0.
FAQ
Comments (12)
I was hoping this would be the Missionary Impossible position where the man is hanging down from a bunkbed to have missionary sex with the woman on the bottom bunk.
You can do it. Prompt for it or have an LLM prompt for that style. You'd be surprised what H3 can do. I had to completely erase constraints I've been conditioned to expect with other models because it was limiting what I could do with H3.
If you're detailed enough, and provide good images, you can get it to do almost anything.
I had an image of my cat sitting on a carpet. Nothing special.
I prompted H3 to send my cat on an adventure to the moon. The video had my cat get up from the carpet, climb into a small shuttle, and get launched to the moon. It landed on the moon, got out, and went "Meow!"
The entire prompt was long and detailed and it followed pretty much everything as written.
So yeah, it can take long ass prompts with complex instructions and make things happen: but, it definitely does best with starting images of some sort.
@DaddyWolfgang Hi wolf, many people say you should use auto prompting or is it enough to use the templates? I write the prompts myself even without templates and it works kinda well but i think the results could even be better. Greetings
Now this.... is a lora! I will be following your work from here on.
Thanks for the buzz! Very kind of you
@LORAGEEK No problem at all. small token
I was slow to catch on... I2VA =D One of the sample has the Workflow.
Oh yeah this definitely improved things without fucking with with the original image. Incredible job!
Tip for others: If you're using images for reference -Start at 0.35 - 0.40 weight. See how it does with the model and other loras you might be using. Adjust from there.
I've been able to get it to do legs together, side missionary (also with legs together), and even doggy and prone is improved.
Great job!!!!
Hey! Would you mind sharing an example of your missionary side-view prompt for H3 Ref2VA? I've tried using Ref2VA with reference images in <Picture 1> and <Picture 2>, but I haven't been able to get it to work properly. I'd really appreciate seeing the basic prompt structure so I can learn from it. Would you be able to send it to me via DM? Thanks a lot! 🙏
@Saxlive yeah, same here please