tokenstore

Photo to NSFW Image: Using Reference Images to Guide the Model

Feed up to nine reference images to the Nureta image API to carry a face, body, or style into an uncensored render — how references work and how to send them.

2 min read

Text alone invents a new person every time. When you want a specific face, body, or look to carry into an uncensored render, you give the image API reference images — up to nine of them — and the model builds on what they show. This guide covers how references drive the render and how to send them.

How references steer the render

A reference image is the source of identity and appearance; the text prompt says what happens. Send a clear reference of a face and the render keeps that face; send several angles and it locks the likeness more firmly. The text still directs pose, framing, and act — the reference just pins who is in the frame.

The content array

The image task takes exactly one text item plus up to nine image_url items. Each image is a hosted HTTPS URL — your own, or the assetUrl from the free upload endpoint.

curl -s -X POST "$TOKENSTORE_URL/api/v3/images/generations/tasks" \
  -H "Authorization: Bearer sk-..." -H "Content-Type: application/json" \
  -d '{
    "model": "seahorse-image", "size": "4K",
    "content": [
      { "type": "text", "text": "same woman, nude, standing by a window, soft morning light" },
      { "type": "image_url", "image_url": { "url": "https://.../face-front.jpg" } },
      { "type": "image_url", "image_url": { "url": "https://.../face-side.jpg" } }
    ]
  }'

Uploading your own photo

No public URL? Ask for a signed upload URL, PUT the bytes, and pass the returned assetUrl as an image_url. Uploads are free — you are billed only on the $0.05 generation. The three-step flow is in the overview docs.

Why send several references

  • More angles of one person = a more stable likeness across the render.
  • Reference a body and a face separately to compose the subject you want.
  • Reuse the same reference set across a whole gallery to keep one identity — see reference images for consistent faces.

Render references at 4K for detail

When the reference carries fine detail you want preserved — skin texture, features — render at size: "4K". The price is flat, so there is no cost reason not to. Distance and framing guidance is in choosing resolution for NSFW images.

Where to go next

New to the endpoint? Start with the NSFW image API quickstart. Building multi-person scenes? See couples and multi-subject NSFW images. Field reference: image API docs.

Try a reference render
Photo to NSFW Image: Using Reference Images to Guide the Model — Nureta Guides