WeShop AIWeShop AI
scroll left

AI Image

Effects

AI Video

Pricing

20% OFF

Solutions

Resource

Download

Affiliate

API

scroll right
Sign In
Resource
Explore guides and tips that help you get things done faster.
blog
Blog
blog
FAQ
feature request
Feature Request
jobs
Jobs
app image
Create Anytime, Anywhere with WeShop AI
Unleash your creativity on-the-go or at your desk. Download WeShop AI and turn inspiration into stunning designs, whenever and wherever it strikes. Powerful, flexible, and always at your fingertips.
October 20

Add a Hidden Ghost to a Travel Photo with GPT Image 2.5

Make a ghost visible in a travel photo with GPT Image 2.5: reframe around a large window, use clear contrast, and follow the tested image-to-image prompt.

A ghost photobomb needs a window you can actually see. This revised example brings the stone guesthouse window close to the traveler and puts a clear white figure behind its glass, so the effect survives a smaller screen.

This AI ghost photo editor workflow starts from one supplied street image. GPT Image 2.5 Sunburst reframes it into a tighter selfie and adds the fictional ghost. The result is a creative Halloween edit, not an unchanged documentary photograph or evidence of a paranormal event.

Reference inputA large window with a visible ghost
Synthetic reference input for hidden ghost photobombGPT Image 2.5 result: a large window with a visible ghost
Tested workflow

One reference image → GPT Image 2.5 Sunburst → hidden ghost photobomb.

Give the window enough space in the composition

Our synthetic reference has a traveler at lower left and a stone guesthouse on the right. In the original wide composition, the windows are small. Instead of hiding a figure in those few pixels, the revised request makes the traveler a head-and-shoulders subject and the haunted window the second main subject. This deliberately changes the framing and building proportions.

Our reference was generated for this demonstration; no real person, pet owner or location is being represented as the subject of a real event. You can use your own suitable photo or another reference you are authorized to edit.

Reproduce the selected result in WeShop

Open WeShop’s GPT Image generator, attach the reference through the image upload control, and select GPT Image 2.5 Sunburst. The source must actually be attached: describing it in words alone would create a different task.

SettingTested value
ToolWeShop GPT Image generator
Modelgpt-image-2.5-sunburst
InputOne attached reference image
Aspect ratio request4:5
Image size / quality1K / medium
QuantityOne per prompt
Selected output dimensions912 Ɨ 1152 pixels

The table reports the requested aspect ratio and the returned dimensions separately. The portrait outputs in this set do not have to be an exact mathematical 4:5 crop. The original attempt and two revisions were generated on October 20, 2026, with one result per prompt. The selected result below uses the final, tighter-composition prompt; this is not a repeat-run reliability test.

Make one window the second main subject

Request a large, front-facing window immediately behind and to the right of the traveler. Give it a rough share of the frame instead of relying on an ambiguous floor number. Preserve the traveler’s recognizable appearance and the stone-house setting while explicitly allowing a tighter camera composition.

Use contrast instead of extreme faintness

Place a pale shrouded head and torso against a nearly black room, with the window frame and mullions in front. Ask for a readable figure at phone width. Our earlier ā€œfaintā€ and ā€œsmall enough to be missedā€ instructions produced a weak reveal; making the window and figure readable is the correction tested here.

Copy the tested large-window ghost prompt

Edit image 1 into a TIGHTER TRAVEL SELFIE composition that makes the haunted window large and obvious. Preserve the recognizable adult man, black hair, glasses, ochre raincoat, old stone guesthouse and blue-hour color atmosphere. Reframe the scene: the man is shown head and shoulders in the LOWER LEFT, occupying roughly the left third; the large dark wooden-framed rectangular guesthouse window immediately behind and to his RIGHT is the second main subject and occupies roughly ONE THIRD OF THE WHOLE IMAGE WIDTH and nearly HALF THE IMAGE HEIGHT. Crop out most of the empty sky and street. The window must be large, front-facing, easy to identify, with a strong stone surround and only a few thin mullions. Inside this one window, behind the glass and mullions, show exactly one clearly visible fictional ghost: pale white shrouded adult head and shoulders, oval face and two dark eye sockets, against a nearly black room. The white head and torso fill about two thirds of the window. Make its shape and contrast readable even when the complete image is shown at 375 pixels wide. A soft cool glow defines the window interior and ghost; preserve warm amber exterior lamps for contrast. The ghost is unmistakable, not a tiny faint reflection. The traveler looks into the camera and does not notice it. Realistic travel snapshot with a playful fictional Halloween photobomb. No extra ghosts, foreground monsters, gore, violence, text, arrows, circles or watermark.

Generate one image, open the result at full size, and compare it with the attached source. The right-hand tool loads this selected example with its actual reference, prompt and settings. You can try it directly using the tool on the right.

What the actual output shows

Revised GPT Image 2.5 travel selfie with a large haunted window and a visible white ghost

The selected revision shows a much larger window beside the traveler, with a white shrouded figure against its dark interior. The head, dark eye area and pale torso are recognizable behind the glass.

The traveler retains recognizable glasses, dark hair and an ochre coat, while the camera composition and house details are regenerated. This is a reference-guided reframe, not a pixel-preserving window-only retouch.

Check the details before you share

The previous published result was too faint at article size. It has been replaced rather than treated as a successful hidden-detail effect. For the replacement, inspect the complete image at the size you will publish: the window surround and white figure should both be identifiable without arrows or a magnifying glass. Architectural geometry and the traveler’s pose can change when the model reframes the image.

If a detail needs correction, return to the source or attach the selected result for a clearly scoped revision. State exactly what should change and what must remain. That correction path is workflow guidance; we are not presenting an untested follow-up as a successful result.

Reveal the window without inventing a ghost sighting

For a carousel, show the original street photo and then the tighter edited selfie. Label the post as AI-created fiction. ā€œI checked the photo after I got homeā€ can be a story premise, but it should not be presented as a factual supernatural report.

Suggested caption:

A perfectly normal travel photo. Until you check the window behind me. šŸ‘» Fictional AI Halloween edit—did you spot it before the reveal?

For a different creative task, see Turn a Bedroom Photo into a Haunted Room with GPT Image 2.5. Start with the loaded example, replace its reference with your own, and adapt one instruction at a time.