Virtual twilight: turn a daytime exterior into a dusk photo

Virtual twilight turns a daytime house photo into a dusk shot with a sunset sky and lit windows. How to make one with AI, and what to check after.

5 min readSume
All posts

Virtual twilight is a real estate photo edit that turns a daytime exterior into a dusk shot: the sky becomes a sunset or deep-blue evening sky, and the windows and exterior lights glow as if the photo had been taken at twilight. With AI, you send the daytime photo to an image-edit model with a prompt for the dusk sky and the lit windows, and say that the house, roofline, and landscaping stay the same.

Below is how to do that with Sume's Image API, whose docs were read on 2026-09-29. Twilight is a relight of the whole scene; Relight a photo with AI covers relighting in general.

What is virtual twilight?

A twilight photo is an exterior shot taken just after sunset, when the sky still has color and the home's lights are on. Getting one for real means a second visit at dusk, with a short window to shoot. Virtual twilight makes that look from a photo taken in daylight: the sky, the light on the walls, and the glow in the windows are changed in editing instead of on location.

Start from a front exterior with a clear view of the sky and the facade, since the evening sky and the lit windows carry the look.

How do I make a virtual twilight photo with AI?

Send the daytime photo as a reference to POST /v1/images with the twilight prompt. The request needs only a few fields:

Request fields for a twilight edit, from the Image API docs, read 2026-09-29.
FieldValue for a twilight editWhat the docs say
modelopenai/gpt-image-2.5ChatGPT Image 2.5 takes up to 16 image references
input_referencesThe daytime photo's public HTTPS URLLocalhost, private-network, and non-HTTPS URLs are rejected before submission
aspect_ratio"auto"On edit calls, auto matches the reference; leaving the field out is not the same
n2 to 4 versions to choose fromUp to 10 per call; per-model ceilings are lower
curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Turn this daytime exterior into a twilight photo: a deep blue evening sky with a warm glow at the horizon, warm light in the front windows, and the existing porch lights on. Keep the house, roofline, windows, driveway, trees, and framing exactly the same.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/listing/front.jpg" } }
    ],
    "aspect_ratio": "auto",
    "n": 3
  }'

What should a virtual twilight prompt say?

Describe the new light, then list what must not move. Useful parts to name:

  • The sky: deep blue with a warm glow at the horizon, or a soft sunset, with or without clouds.
  • The windows: which ones glow, and in what color of light. Say “the existing windows” so no new ones appear.
  • Exterior lights: only fixtures the house really has, such as porch lights or path lights.
  • What stays: the house, roofline, windows, doors, landscaping, driveway, and the framing.
  • The ground and walls: ask for them to darken to match the evening sky, so the scene doesn't look like daylight under a dusk sky.

Will the house look exactly the same?

Not guaranteed. The model draws a new image, so window shapes, trim, plants, and fixtures can drift, and it can add lights the home doesn't have. Compare each version with the daytime original before you use it, and keep the original.

Sume's Terms of Service make you responsible for reviewing outputs before publication, and bar using the service to mislead people about the authenticity of generated media. Whether a listing must say a twilight photo is edited depends on your platform and local rules; check them.

How much does a virtual twilight cost?

A virtual twilight is an ordinary image edit, billed per completed image at the model's endpoint price, and a failed or cancelled generation is not billed. The price lookup, the agent fee, and the other limits are the same as for any listing edit; see AI real estate photo editing.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume