Prompt field guide ·
GPT Image 2.5 prompts: image editing and ecommerce consistency
Original copyable prompts for model outfit and scene changes, multi-reference editing, product cutaways, documentary photography and three-panel stories, with review checklists.
Write an edit as a contract
- Assign references: A controls identity and composition; B controls the garment or product; C controls only the environment. Use images you have permission to use.
- Name the invariants first: identity, pose, framing, camera height, subject scale and contact points. Then describe the one change you want.
- Generate one edit and inspect it against the original before adding more changes. For a failure, correct the specific defect instead of rewriting the whole prompt.
- For complex edits, separate garment replacement from background replacement. Feed the accepted garment edit into the background step while retaining the original identity reference.
01 / Same-model outfit and scene variations
Create comparable apparel campaign variants from one approved full-body model reference.
Reusable editing template: attach the approved model photograph and replace the outfit and scene placeholders. The actual downloaded sequence is shown below.
Original generation prompt (English)
Edit the approved model photograph into a tailored cobalt-blue trouser suit over an ivory crew-neck T-shirt in a contemporary pale-stone gallery. Preserve identity, expression, head center, shoulder line, hands, hips, knees, subject scale and position, camera, crop and the original white shoes. Fit the clothing to the existing pose. Adjust only the environment and the contact shadow needed for believable placement. No re-posing, extra accessories, text or collage.Review: No. The prompt requests stable identity, pose and placement, but generation can still move or redraw details. Compare facial landmarks, wrists and shoe tips before accepting each frame.
Open this case →02 / Labeled espresso machine cutaway
Use for a conceptual product explainer with separate flow paths and a short, exact label set.
Original generation prompt (English)
Design a portrait educational cutaway of a fictional automatic espresso machine on an ivory background. Separate the bean route from the water route using brown and blue arrows. Include exactly these six labels: "BEANS", "GRINDER", "WATER TANK", "HEATER", "BREW GROUP", "CUP". Show a readable left-to-right brewing sequence, copper mechanisms and restrained teal accents. No brand marks.Review: Check every label and arrow before teaching from the image. This fictional cutaway is a visual concept, not an engineering specification.
Open this case →03 / Miniature moon repair workshop
Use for handmade fantasy key art, tactile miniature scenes, and contrasting warm and cool light.
Original generation prompt (English)
Create a square photograph of a handmade miniature observatory workshop. An adult craftswoman in a navy apron uses a fine brass tool to mend a luminous crescent moon on her workbench. Tiny star-shaped spare parts sit in labeled-free wooden trays. Emphasize felt sleeves, imperfect clay craters and brushed copper. Amber desk light meets blue starlight through a circular window. Clear hands, quiet concentration, no lettering or watermark.Review: Inspect the tool grip, hand anatomy, crescent silhouette, and separation between the subject and workbench.
Open this case →04 / Harbor craft documentary portrait
Use for fictional documentary-style portraits that emphasize age, craft, and natural material detail.
Original generation prompt (English)
Photograph a fictional elderly fisherwoman repairing a rope basket on a harbor quay, with a small brown dog beside her boots. Eye-level documentary framing, overcast dawn, faded ochre raincoat, salt-worn wood, natural wrinkles and realistic hands. Leave breathing room around both subjects. Understated film color, no beauty smoothing, text or logos.Review: Inspect hands, basket weave, dog anatomy, and the crop. The subject is fictional; this is not a photograph documenting a real person or event.
Open this case →05 / Three-panel paper boat cat story
Use for a short visual narrative with one action per panel and consistent recurring characters.
Original generation prompt (English)
Draw a landscape comic with exactly three equal panels and narrow cream gutters. A curly-haired adult man in mustard overalls shares a studio with a black cat wearing a red collar. Left: he folds a paper boat. Middle: the cat taps the boat into a shallow blue tray. Right: both watch it float. Keep clothing, face and collar consistent. Textured gouache, cobalt and honey palette, expressive poses, no speech bubbles or lettering.Review: Check that there are exactly three panels and that the face, overalls, black fur, and red collar stay consistent.
Open this case →Build a same-position ecommerce sequence
Start with one accepted full-body anchor. Record the aspect ratio and approximate head, waist and foot coordinates. Every new outfit should branch from that anchor; repeatedly editing the previous result can accumulate identity and composition drift.
- Actual frame order: white tee and charcoal trousers in a gray studio → cobalt-blue suit in a gallery → terracotta knit and beige trousers in an arcade → green trench in a greenhouse. The AVIF loops these four downloaded stills, holding each for two seconds.
- Add to each edit: Keep head center, shoulder line, hands and both feet at the same image coordinates as A. No zoom, pan, rotation or crop. Change only the specified garment and environment.
- Compare every frame at the same display size. Reject identity changes and visible position shifts; a prompt is a constraint, not a guarantee of pixel-perfect registration.
- Download the accepted originals from the web interface. Preserve frame order and equal dimensions, then encode a looping animated AVIF with enough hold time to compare each outfit. Provide still frames and a reduced-motion alternative alongside the animation.
A try-on image is a visual concept: verify fit, fabric, print placement and product details against the real item before using it in a product listing.
Transparent model cutout
Original generation prompt (English)
Remove the background of the approved full-body model photograph. Keep the original head, fine hair edges, face, pose, outfit, hands and white shoes at the same scale and frame position. Remove the ground shadow and make the gaps around the body genuinely transparent. Return one RGBA PNG; do not draw a checkerboard or a white background.More reference-image workflows
Further reading & provenance
The prompts on this page are original editorial examples. These resources informed the workflow; community examples are references, not independent verification of every claimed result.
Generate with GPT Image 2.5
Or pick a style to start from:
206Image resolution
Free generation uses 1K. An active subscription supports 2K and 4K generation using credits.