Generating Multi-Subject and Couples NSFW Scenes as Stills

One person in a frame is easy — the model has a single body to place and light. Two or more is where prompts get sloppy, bodies blur together, and it stops being clear who is doing what to whom. The fix isn't a special mode; it's describing the arrangement precisely and giving the model a reference for each person you want to keep consistent.

This guide covers rendering couples and multi-subject scenes as single stills on seahorse-image: how to lay out positioning and interaction in the prompt, how to pin each subject with reference images, and how to keep everyone on-model in the same frame.

Describe the arrangement#

State who is where in physical terms the model can place. Anchor each person to a position and to each other: who is in front, who is behind, who is above, which way each faces, where hands and hips are. a couple in bed, the man behind the woman, both facing the camera, his hands on her hips is renderable. a couple being intimate is not — the model has to invent every one of those relationships, and it invents inconsistently.

Name the interaction#

The single most important word in a multi-subject prompt is the verb connecting the subjects. Explicit acts render directly, so say the act plainly and attach it to the specific bodies: who is doing what to whom, in which position. Vague coupling produces two people who happen to be near each other; a named act produces an actual scene. The prompt structure that works for one subject applies here too — see writing prompts for explicit AI images — you're just doing it per person and per relationship.

Framing for more than one body#

Multiple bodies need room and a clear vantage. Wider framing and a stated angle help the model keep everyone whole instead of merging limbs. wide low-angle shot or side-on full-body shot gives space for two or three figures; an unspecified tight crop is where extra arms and fused torsos creep in. If you want an explicit close-up of the point of contact, render it at higher resolution so the detail holds.

One reference per subject#

To keep each person recognizable, give the model a reference for each. An image task takes up to nine image_url references, so a couple can have two or three references apiece and still fit. In the prompt, tie each described person to their look so the model knows which reference maps to which body — for example distinguish them by hair or a tattoo you also mention. The mechanics of how identity and markings carry are in using reference images for consistent faces and bodies.

curl -s -X POST "$TOKENSTORE_URL/api/v3/images/generations/tasks" \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seahorse-image",
    "content": [
      { "type": "text", "text": "a nude couple on a bed, the tattooed man kneeling behind the red-haired woman, his hands on her hips, both facing the camera, wide low-angle shot, warm lamplight" },
      { "type": "image_url", "image_url": {"url": "https://.../man.jpg"} },
      { "type": "image_url", "image_url": {"url": "https://.../woman.jpg"} }
    ],
    "size": "4K"
  }'

If your references aren't hosted yet, upload each once via the presign flow and reuse the returned URLs across the whole set.

Keeping both consistent across a set#

For a series featuring the same pair, lock both reference sets and reuse them in every request, changing only the prompt's arrangement and act. That's how a couple stays the same two people across a dozen shots instead of subtly swapping faces frame to frame. Keep references sharp and give each subject a distinguishing feature you can name — it prevents the model from blending them into one look.

Where to go next#

• Lock each person first: using reference images for consistent faces and bodies.
• Build reusable subjects: building an NSFW character sheet with AI.
• Get the anatomy right at the point of contact: getting anatomy-accurate nude art from AI.