Oral Creampie — CIMLG v2 (MiniMax H3)
He finishes in her mouth, and you actually see it happen. His cock pulses, the urethra on the underside swells and empties with each pulse, and cum spills out over her lips and runs down the shaft.
This is the oral creampie / cum-in-mouth finish. He stays in her mouth while he cums, and the pulsing is the whole point.
What this LoRA is (and isn't)
This is an augmenter, not a motion LoRA. It adds the pulse, the throb and the cum-in-mouth finish to your scene. It brings some motion with it, but the blowjob itself (her head moving, his hips) comes from your prompt and your base model. Pair it with a motion LoRA if you want more movement (see Motion Booster below).
No cum LoRAs needed. The cum in this LoRA looks real on its own: white, wet, semi-transparent, and it flows out of her mouth naturally. Every example on this page uses no cum add-ons unless it says so.
Image to video (I2V). It was trained as I2V, so it starts cleanly from your start image with no frozen first second and no camera jump.
Quick start
Base model: Eros Max beta6 (turbo built in, 8 steps)
Strength: 0.4 (range 0.4–0.5)
Length: 10 seconds
Trigger:
CUMOUF(optional)Motion Booster V2: 0.5–0.8 (optional)
Sound: simple, see the sound section
⚠ Never run it at 1.0. At 1.0 you get deformed anatomy, strange cuts in the shot, the urethra splitting open like a slit, and red cum that looks like blood. 0.4 is the sweet spot. 0.7 brings back a chewing jaw.
Stock H3 is not recommended. On stock H3 (without Eros) she tends to chew on it. Eros Max beta6 gives a real blowjob.
10 seconds is the sweet spot for a full act: sucking, then the finish. 5 seconds works, but everything happens too fast. For a short clip, prompt just the finish.
The trigger word
CUMOUF is optional. The LoRA works as soon as it is loaded. The trigger adds a small push.
Put it at the very start of the prompt, before everything else.
Never put it inside the action text, and never put a colon after it (
CUMOUF: she...). H3 reads a word followed by a colon as a speaker's name, and she will say "come mouth" out loud. Some users find she says it even at the start. If that happens, just leave it out.
The training words
The LoRA learned from 412 clips, each captioned with the same three parts: shot size, camera view, then the action. Use these exact words. They are what the LoRA responds to best.
Shot sizes
Close-up (the most training, use this first)
Extreme close-up
Medium shot
Wide shot (very little training)
Camera views
Filmed from the side (the most training)
Filmed from between his legs, looking at the underside of his cock (where the pulse shows best)
Filmed from above, looking down (his point of view)
Filmed from below, looking up
Filmed head-on at the tip of his cock
Filmed facing her mouth
Write your camera line like the training: Close-up. Filmed from between his legs, looking at the underside of his cock.
The actions, in order of the act
1. Entry (soft or starting):
His soft cock slides into her mouth.
2. Thrusting:
He holds her head still and his hips push forward, thrusting his cock in and out of her mouth.
3. The pulse (the heart of this LoRA, by far the most training):
His cock pulses in her mouth as he cums, the urethra on the underside swelling and shrinking with each pulse.
Short version:
His cock pulses in her mouth as he cums.
4. Spurts into an open mouth:
Cum shoots out of the tip of his cock in arcing spurts into her open mouth.
5. Spilling out:
Cum spills out of her lips.
6. After:
Her mouth is open with cum on her tongue.
She opens her mouth to show the cum on her tongue, then swallows.
Writing the prompt
Make her the one moving. Write "She sucks his cock", not "her lips slide down his cock". When her lips or mouth are the subject of the sentence, H3 animates her jaw instead of her head, and she looks like she's chewing.
Don't park her mouth. Lines like "her lips hold still" or "she holds her mouth closed over the head" also turn into a chewing jaw. If you want stillness, put it on him: "he holds still" or "he holds her head still".
Put the cum early. H3 plays your prompt in order. If sucking comes first and the cum last, the cum lands in the final second and the clip ends just as it gets good. Keep the build-up short, or open with the pulse and let her keep sucking while he cums.
One shot. Every extra shot or scene is a camera cut, and cuts eat into the cum time. Keep the whole act in one shot.
Extra actions after the cum. If you want something after the finish (she swallows, smiles, he touches her), put it at the end of the same shot. Make sure it fits your picture: if you ask him to squeeze something that isn't in reach, he'll grab whatever is nearest.
Match your start image. Describe what's actually in the picture: where they are, who's where, what they're wearing. If your setting words say "car" and the image shows a theater, H3 moves them.
Example: full act, 10 seconds
CUMOUF. Close-up. Filmed from between his legs, looking at the underside of his cock.
She lowers her mouth onto his cock sliding it up and down. She keeps his cock in her mouth and sucks it slowly, closing her eyes. His cock pulses in her mouth as he cums, the urethra on the underside swelling and shrinking with each pulse. Cum spills out of her lips. She swallows, looks at him and smiles.Example: cum right away, 10 seconds
CUMOUF. Close-up. Filmed from the side.
His cock pulses in her mouth as he cums, the urethra on the underside swelling and shrinking with each pulse, while she sucks his cock slowly and looks up at him. Cum spills out of her lips.Example: open mouth, tongue out
CUMOUF. Close-up. Filmed from between his legs, looking at the underside of his cock.
Cum shoots out of the tip of his cock in arcing spurts into her open mouth. She takes his cock into her mouth and sucks it slowly, looking up at him. His cock pulses in her mouth as he cums, the urethra on the underside swelling and shrinking with each pulse. Cum spills out of her lips.Sound: keep it simple
This LoRA was trained without audio, so all the sound comes from your base model. That makes your sound words very powerful: they change the whole video, not just the audio. A sound needs a matching movement, so H3 makes one.
Start with this and add one thing at a time:
Quiet ambient room sounds. Soft sliding sounds, wet skin gliding on skin. He groans as he cums.
Avoid sucking, slurping, gagging, coughing or eating sounds. H3 makes her work her jaw to produce them, and it sounds like crunching celery.
Avoid places that sound like eating. "Movie theater sounds" brings popcorn crunching with it. Use "quiet ambient room sounds" and describe the place in the picture instead.
Don't leave the sound empty. H3 fills silence with invented voices, sometimes a whole crowd.
Outdoors: "Quiet outdoor ambience" instead of room sounds.
Music: describe the instruments and mood ("playful pizzicato strings and soft piano"), not "a movie soundtrack".
Longer scenes: two clips, joined
H3 gets messy past about 12–15 seconds: people change faces, clothes change color, things jump. For a longer story, make two clips:
Clip 1 (about 10–12 s): the blowjob and the cum.
Clip 2 (about 8 s): what happens after. Use the last frame of clip 1 as the start image.
Join them in any video editor.
Describe the people the same way in both prompts (age, hair, clothes) so they stay the same.
Recommended: my Cum Slider for volume control
CUM : The LG Cum Slider (H3). Strength is the dial, no trigger word.
Negative strength = less cum. Around -1.33 gives a natural amount.
Around 1.0 the cum turns from a nice, semi-transparent look to a thicker, whiter cum.
Above 1.0 gets silly. If you want a flood, see the slider page.
Write "white cum" in your prompt, or it can come out pink.
Recommended: Beatbox to build the prompt
LoraGeek Beatbox - Prompt Builder + Camera Panel. Free. Type your beats one per line, click your camera, and it writes H3's full prompt layout for you, with cut times and dialogue tags done right.
Put
CUMOUFin its trigger word box. It goes at the very start, never as speech.Each line in a shot box is a beat in the same shot. Each new shot box is a camera cut.
Whispers stay whispers: write
Granny whispers: "Good job."
Optional: fixing the penis angle
In some angles H3 flips the head of the cock toward the camera. If that happens, stack HMPenis by HearmemanAI at about 0.45–0.6. Most clips don't need it.
Known limits
Stock H3 (without Eros): she chews.
The penis head can flip toward the camera in some angles.
The pulse shows best in the between-the-legs view, and in close-ups.
Medium and wide shots show the cum, but the pulse is small.
How it was made
25 hand-picked close-up and macro videos, every one narrated by hand to mark exactly where the pulse, the spurts and the spill happen.
412 training clips, captioned shot size → camera view → action. The best, most visible pulse moments counted double.
Trained as image to video (I2V), audio off. Released at step 2500 of 4000.
Older versions have their own notes on their tabs.
Description
Quick start
Update: Verified works well with LTX 2.5
Base model 10Eros v1.5 Trigger CUMOUF — first word Strength 1.0 CFG 1.2 – 1.3 Length 8 – 15 seconds Distilled LoRA ltx-2.3-22b-distilled-lora-384-1.1 Sigmas the longer schedule below — this matters
First-pass sigmas. Worth using: switching to this schedule improved the spasm, the detail and the audio noticeably over a shorter one.
1.000, 0.955, 0.893, 0.812, 0.715, 0.603, 0.482, 0.241, 0.121, 0.0
The prompt used for the example renders:
Performance: [CUMOUF. Close side view of his cock in her mouth, her lips wrapped
tight around the shaft. He pumps slowly into her mouth, then holds deep as he starts
to cum. You see his cock twitch and spasm over and over, arching and pulsing as it
pumps into her. Thick white cum fills her mouth and spills out over her lips, sliding
down the shaft.]
Sounds: [Man grunts loudly as he cums. Woman coughs and sputters trying to swallow it all]
Dialog: [(off camera) Man says: Take my load in your mouth. Swallow my cum.]
Note the three separate blocks — that matters, see below.
Three things that will waste your time if you don't know them
1. Don't ask for a man's voice without saying he's off-camera.
LTX-2.3 lip-syncs. Ask for a line and it needs a face to put it on — so it drags one into your shot, and once there's a second mouth near a cock it does the obvious thing with it. Cost me an hour to work out.
Dialog: [(out of frame) Man says: swallow it]
2. Sound goes in its own block, or it gets spoken out loud.
I wrote "he grunts loudly" into the visual description and got a render where the man said the words "he grunts loudly". A grunt is a facial action, so describing it in the picture asks for a face doing it.
Performance: [ the action — no sounds, no speech ]
Sounds: [ wet mouth sounds, grunting from the man, moans from the woman ]
Dialog: [ (out of frame) Man says: ... ]
3. pumps means ejaculating here, not thrusting.
It's in 82% of the training captions as "as it pumps" — the cock pumping cum. So "he pumps fast in and out" asks for the climax while you're trying to ask for the build-up. Use thrusts for motion.
Words it knows
Measured across all 1,525 captions:
word in cock 100% spasm 79% contract 79% hard (intensity) 48%
Zero occurrences — these fall through to the base model: penis, dick, load, spasms (with an -s), twitches, jerks, arch.
That last group matters. spasm and spasms are different tokens. The trained phrase is "spasm and contract over and over" — use it as written.
It moves in for the finish — that's the point
Start from a normal, wider shot. When the finish begins, the camera pushes in on its own.
That is the behaviour, not a side effect. It's a finisher, and this is what a finisher should do — the same move an editor would make. You stay in the scene through the act, and when it matters the shot tightens to show you the thing you came for.
There's a physical reason it has to. A spasm is a few millimetres of movement. From any distance it's sub-pixel and simply doesn't exist on screen — you can't see one from across the room. Every clip I trained on was cut tight for exactly that reason, so the model composes the same way: it frames for its subject.
To get more of the move: start wider, and state the spasm phrase twice. To get less: start closer, and state it once. A start image already cropped tight gives it nowhere to go, and puts more pixels on the spasm from frame one.
Your start frame decides what his cock looks like
Raised in the comments, and it's a good catch — thank you.
The model only knows the shape of something from what it can see in frame one. If it's already halfway into her mouth, the head was never visible, so when she comes off him it has to invent one — and it invents an ordinary one. If yours is distinctive at all, that's where the detail goes.
So start with it near her mouth, not yet in it. One clean look at the shaft and head is enough and it holds for the rest of the clip.
The trade-off: a frame already mid-act starts moving sooner, because this continues an action far better than it starts one. About-to-go-in is the sweet spot — clear anatomy, and something already happening.
It will pull back at the end. That's trained in, not a fault: a lot of the dataset is the finish continuing after she comes off him, cum running out of her mouth and down the shaft. To push against it, keep the render at 8 seconds so it ends during the finish, write that she holds him deep and swallows, and say nothing about it spilling down the shaft. None of that is reliable — showing it clearly in frame one is.
To get it out of her mouth: start from a frame where he's already out, or say she is looking at it and holding still.
Strength — 1.0, and the range is narrow
Alpha equals rank, so 1.0 is genuinely the LoRA as trained.
0.7 clean but flat 1.0 use this 1.5 stronger, but body morphing appears
If the spasm is weak at 1.0, more strength won't fix it — that's a data limit.
Known limits
The audio is trained, not the base model's. It generates its own sound — wet mouth sounds, breathing, and grunts that came from the source footage rather than the gritty default. Use the longer sigma schedule; it makes a real difference to the audio too.
It's a finisher. 1,169 of 1,525 clips are the climax itself; only 93 are in-and-out motion. It won't drive face-fucking or a long build-up — use another LoRA for the act and bring this one in for the finish.
Anatomy holds up but isn't perfect. Nine continuous seconds in her mouth and it comes back recognisably the same — that used to render as chewed-up meat. It still stretches sometimes on the way out, and it can only keep what your start frame actually showed it.
Long renders drift. Trained on 2-second clips. 8–15 seconds is the tested band.
LTX training
Base 10Eros v1.5 bf16 Clips 1,525 × 49 frames @ 25 fps Sources ~14 couples plus solo footage, every window timed by hand Bucket 640×384×49, rank 8 / alpha 8 Branches video + audio, trained jointly Hardware one RTX 5090 Steps trained to 12,000; released checkpoint is step 11,000
Later is not automatically better. I tested the whole ladder and 11,000 was the best trade between spasm strength and stable anatomy — the last checkpoint was not the one I shipped.
FAQ
Comments (20)
Nice job. This made me stop using MiniMax for a second. It works well. I posted a video but it was flagged for review so I guess will see it later. LTX2.3 definitely won't go away.
Ill look at doing same thing for MiniMax. Thanks!
i know, right????????????????????????????????? I think this is the first oral creampie on Civit AI, actually congrats!!
@The_Last_Goblin_King I made it for that reason. I couldnt fine one myself :) Thank you!
Despite what you or other think. LTX2.3 ad Wan22 will remain the main model for NSFW content. for one simple reason, it's a lot faster. Even with Lightx2v lora, H3 is slow for genrating prduction ready HD videos.
So, nohing will change this in the futur, LTX2.3 and Wan22 will remain the faster way to genrate HD NSFW content for PAtreon, OFM PPV.
not to mention, that native Wan22 face/body consistency reamain unbeaten for now.
@Pat3dx I used LTX2.3 exclusively. I haven't used it all except to make this lora's video since Minimax was released. The only time I'll use LTX2.3 is to trying out interesting Loras. I don't need to debate it though. Just look at stablediffusion reddit to see what's preferred. MiniMax videos posted number about 100 to 1 vs LTX2.3. When a good model arrives on the scene people use is (Like Krea2). Why debate though? if LTX2.5 trumps MiniMax I'll use it. I've made 10's of thousands of videos with both Wan and LTX. (hunyuan before that). MiniMax is on another level. I have no "hate" for any model. I use what I like as should you. Wan still has some kinky fetish stuff not out on MiniMax yet.I support all open weight models but my god MiniMax is good. I'm looking forward to the next LTX release.
@iodrg244 Another level is an understatement. This is game-changing.
@Pat3dx Lol? I guess you haven't tried H3 enough...
@GlowingGuardianGirl lol not enough. I used it since it's been released to make my cutom node for this new model and workflow. Trust me i tested a lotttt.
Go try to generate an H3 native 20-30s HD video with int8 conv+ lightx2v 4 step lora at 8-10 steps, lower it's not publishable for subscribers, which is the only model for the most relevant production ready quality to publish for subscribers. and try to make the same LTX2.3 or even 2.5, as i used both now, same 20-30s video, and you will see H3 is much longer.
In the same generation time, i can make a 1080p LTX2.3 or 2.5 video while H3 generate an HD video.
@Pat3dx Made my day 😂😂😂😂😂😂🤣
same for H3 please 🙏🙏🙏🙏🙏
I’ll release it tonight! On the last bit of training now
@LORAGEEK 😍😍😍😍 you are the goat! doing the lords work right here!
Hey Azra, its out. Check the writeup for how to use the H3 version vs LTX. Weird they cant have same page but different "tabs"
@LORAGEEK hey, yes i already downloaded. will definetly try it out! thanks a ton, you da best!
Great lora! hope you can make one for a creampie as well
Very good lora, well done :).
The only issue i noticed, is that even when the start frame the cock is already in her mouth, at the end, she move her head back and the cock comes out.
This is a critical issue, as LTX don't know what was the shape and lengh of the cock, the glans is very bad.
Example, my first frame the cock is already in her mouth, this is a futa huge black cock, and half of the cock is in the mouth.
At the end, she moves her head back, cock comes out, but the cock is no longer a huge black cock.
So, my advice, never start with a frame with the cock already inside the mouth, if you want perfect consistency with the shape of the cock/glans.
Start with the cock, near her mouth and not yet inside, that way LTX2.3 perfectly know the shape/lengh of the cock, especially the glans.
And at the end when the girl move back her head and cock comes out, it will be perfect consistency.
Or is there a way to make the girl at the end, don't move back her head, and keep the entire cock inside her mouth?
Thanks and that's a really good catch. You're right about the cause.
LTX only knows the shape of something from what it can actually see in frame one. If half of it is already in her mouth, it never saw the glans, so when she pulls back it has to invent one and it invents an average one, not yours. Nothing about the start frame tells it "huge and black" if the huge black part is hidden.
So your advice is right and I'll put it on the page: start with it near her mouth, not in it. One clean look is all it needs and it stays consistent the whole way through.
On keeping it in her mouth at the end the pull-back is partly trained in. A good chunk of the dataset is the cum running out of her mouth and down the shaft, so it wants to end with it visible. You can push back against it:
- shorter render (8s) so it ends during the finish rather than after it
- say she holds him deep in her mouth and swallows, and leave out anything about it spilling down the shaft
- drop the strength a little
But your own fix will give you better consistency than any of that. I'm noting it for v2 a group of clips where she stays down through the whole finish would make "stay in" an actual thing you can ask for.
@LORAGEEK yes, this is not a particular issue specific with LTX2.3, wan 22 and H3 have the same issue, and i think, all futur model will also have this issue, it's a normal issue. Model can't know something it can't see.
But my first test where the girl have half of the glans inside her mouth, was a success. Even at the end when she pull her head back, the consistency and glans shape was almost perfect. :)
Check it, i just uploaded the video on your page.
@LORAGEEK thanks for the H3 version, will test it this afternoon :)
Details
Files
CUMOUF_oral_creampie_v1.safetensors
Mirrors
CUMOUF - Oral Creampie - LTX2.3 - cfg1.2-1.3,8-15seconds.safetensors
CUMOUF - Oral Creampie - LTX2.3 - cfg1.2-1.3,8-15seconds.safetensors
CUMOUF - Oral Creampie - LTX2.3 - cfg1.2-1.3,8-15seconds.safetensors
CUMOUF_oral_creampie_v1.safetensors
CUMOUF_oral_creampie_v1.safetensors
CUMOUF_oral_creampie_v1.safetensors