First threesome prompt I ever ran gave me a woman with two heads and a guy with three arms. The prompt was not the problem. The model gave up somewhere around body number two.
Group sex prompts break for one mechanical reason. Every model degrades as you add people. Faces converge until everyone is the same person. Limbs get reassigned to whoever is nearest. Torsos fuse at the hip. Prettier adjectives will not save you. Structure will. Count first, identities second, contact third, negatives last.
Lead with the number or lose
The model commits to a composition in the first few tokens. If “three people” shows up in word 40, it already built the shot around one subject. So the count goes first, as a numeral. “3 adults, two women and one man, explicit.” Then never contradict it. Say three and later mention “the other guy” and you quietly asked for four. You get four, and the extra one is built out of leftovers.
Common fail
Everybody in frame has the same face, same hair, same body. A clone orgy.
Fix: every person gets a full identity block, not one adjective. Hair color and length, skin tone, body type, and one hard anchor like a choker, glasses, a tattoo. The model cannot reuse one face and satisfy all three.
Example prompt
“3 adults, explicit, one man and two women. Woman A: tall, pale skin, short platinum bob, black choker. Woman B: shorter, deep brown skin, long black curls, gold hoops. Man: mid thirties, olive skin, dark stubble, lean build. Woman A kneels on the bed facing camera, the man kneels behind her, Woman B sits to her left. Warm lamp light, shallow depth of field, 35mm, waist up crop.”
Assign every hand an owner
This is where group prompts die. “Two women touching a man” has no owners in it, so the model invents anatomy to satisfy the sentence. Write contact as owner, limb, action, target. “Woman B’s right hand on the man’s chest.” “The man’s left hand in Woman A’s hair.” Boring. Works.
Cap it at three or four contact points. Past that the model sprouts limbs to cover the list. For act and position vocabulary models respond to, the NSFW prompt dictionary has the terms.
Common fail
A fifth hand drifting in from nowhere, or a leg attached to the wrong person.
Fix: name the limb and its owner, then put “no extra limbs, no extra hands” in negatives. Drop any body part you did not deliberately place. Never mention feet, less to guess at.
Stage in depth, not in a pile
Left, right and center will not cut it. Bodies at the same distance from the lens merge into one mass. Put people on different planes and different heights. One kneeling in the foreground, one standing mid frame, one seated behind. Add “bodies clearly separated, visible gap between torsos, no overlapping figures” and let shallow focus finish the job. A slightly soft back body reads as a person, not a growth.
Example prompt
“4 adults, explicit group scene, one woman and three men, dim club back room. Woman in foreground on a low bench, red hair, freckles, facing camera. Man 1 standing right, dark skin, shirtless, face turned away. Man 2 standing left, tattooed forearms, cropped at the chin. Man 3 behind him, out of focus. No overlapping torsos. Low key red light, 50mm, chest up on the woman.”
Framing carries more than the words
A wide shot asks for four complete anatomies and you get four broken ones. Crop instead. Waist up, over the shoulder, or a tight two person frame with the third body entering from the edge. Every face you keep out of frame is a face that cannot go wrong.
Example prompt
“2 adults in frame, third person implied. Close two shot, blonde woman and brunette woman, faces close together, explicit, a man’s hands entering frame on the blonde’s waist, his body out of frame. Soft window light, 85mm, shallow focus, bedroom.”
The negative block that matters
- extra limbs, extra hands, extra fingers, three arms
- duplicate face, identical faces, cloned people
- fused bodies, merged torsos, conjoined
- floating limb, disembodied hand
- extra person, crowd, background people
- deformed anatomy, melted hands
“extra person” earns its spot. Group prompts leak spare bodies, and one half rendered stranger ruins the shot.
Where the ceiling sits
Three holds. Four is a coin flip. Five or more and nothing on the market renders it cleanly. Anyone selling you a nine person gangbang prompt is selling a lottery ticket.
So cheat the count. Crop until two or three bodies are fully visible and let hands, thighs and out of focus shapes imply the rest. Keep one face sharp, turn or crop the others. Of everything in my tool rankings, ClothOff handles multiple people best.
All of this rests on a habit from further back. Once the headcount is honest the placement still has to be described body by body, never named and hoped for, and names mean nothing, geometry does is where that discipline comes from.
Two heads and three arms is the model guessing at a crowd. Give it a headcount it can hold.
