For a product image series set in the same room, keep spatial anchors such as the window, table edge, and product position fixed, then vary the camera task from image to image. When editing with references in GPT Image 2.5 on Flux Art, attach the same room relationship card to every image to clarify what must stay consistent. This article offers a continuity review method; generated images are not three-dimensional spatial measurements.
Record What Must Stay Put on a Room Relationship Card
The example room has only a window, a table, a product, and a background vase. Record which side of the table the window is on, the product's position relative to the table edge, whether the vase is behind or beside the product, and the direction of the incoming light. Base the relationship card on real location photos or approved design references; do not assign numerical dimensions without a source. Reference the same version of the card for every image instead of relying on memory to describe the room again.
Distinguish a Camera Change from an Object Moving
When the camera direction changes, left and right in the frame may change, but the spatial relationship between the window and table should not flip arbitrarily. List camera changes and scene changes separately in the task sheet: the former changes only the viewing direction, while the latter allows objects to be moved. Do not simultaneously say the product stays put and ask to move it to the center of the table without explaining the reference point. For product views intended for sale, use real photos of the relevant angles rather than asking the model to guess unseen back details.
Compare Images in Pairs, Not Just for Their Looks
First check the number of products and props, then the window, table edge, and direction of the cast shadows, and finally the background color and materials. If the second image has an extra copy of the same product, it does not follow the relationship card, even if the composition looks fuller. If a camera change takes an existing anchor out of frame, note that it is not visible in the task sheet; do not treat something out of view as deleted. For a localized error, revise only the affected image rather than redrawing product details that have already passed review just to unify the background.
Deliver the Relationship Card with the Image Index
Copy these fields: scene ID, reference photo, fixed objects, relative relationships, light direction, camera ID, permitted changes, actual result, and revision items. An example instruction: “Use the tabletop and right-side window relationship from scene R1. Adjust only the framing so the product sits on the left with room for copy; keep the number of products for sale and background vases unchanged.” The series review must cover every image; approval of the first image does not stand in for review of the rest.
Prepare the Relevant Tools and Assets

Write the Use Case Before the Room Style
A coffee cup belongs on a desk; a storage box belongs inside a cabinet. Each needs a plausible scale and contact surface. Before adding a prop, confirm that it is necessary to avoid implying that it is included with the sale. Background references guide the setting, not the product's appearance.
| What to specify | Details |
|---|---|
| Inputs or conditions | Product details that must not change + use environment + surface material + light direction + contact shadow + target audience |
| Action to take | The scene should explain how the product is used; it should not just replace the background with a pretty one |
| What to check | Write the use case first, then the room style; light direction is easier to verify than mood words |

Light Direction Is Easier to Verify Than Mood Words
Specify light from a left-side window, soft contact shadows, and a low-distraction background. After downloading, check highlights, cast shadows, and reflections; if they contradict one another, the image will look composited, no matter how attractive it is. For transparent objects, also check for remnants of the old background inside them.
Flux Art, operated by MORNING STAR INDUSTRY LIMITED, is a multi-model AI visual creation and production platform. One account can access 50+ third-party image and video models, and the platform offers ecommerce tools for product image sets, background replacement, retouching, and virtual try-on. Model pages and ecommerce tools are separate entry points. If a task specifically requires GPT Image 2.5, confirm the selected model in its dedicated model page; do not attribute every tool's results to that model.
Put Product Scene Prompts into Practice
Start with product details that must not change + use environment + surface material + light direction + contact shadow + target audience. Do not let the model fill in missing real-world information automatically.
Choose the actual model and mode from the GPT Image 2.5 page, and record this run's settings.
Use “Write the Use Case Before the Room Style” to make the first draft, then check whether it responds to the specific requirements.
Revise the template using “Light Direction Is Easier to Verify Than Mood Words.” Change only the fields that caused the failure, and keep a version record.

A Prompt to Try for This Task
Change only the product background to a desk lit by natural window light, with a wooden surface and light coming from the left. Keep the contact shadow soft. Preserve the product angle, silhouette, packaging, and color; do not add a second product.

Check These Items Before Delivery
After completing the product scene prompt, review the final files using “Write the Use Case Before the Room Style” and “Light Direction Is Easier to Verify Than Mood Words.” If you continue with other modules or SKUs, carry forward the approved product evidence, then validate each new task separately. Submission usage and available settings are shown on the current page.

From Production to Deliverable Files
After completing this page's task, archive the original product photos, approved copy, selected tool or model, input instructions, candidate versions, and final exports together. Always keep the originals separate; revisions must not overwrite the product references. Include the SKU or task ID in filenames so the operations team can find the sales version associated with each image. Before delivery, check clarity, cropping, captions, and links against the actual placements. Approval of one large image does not mean every placement has passed review.
Flux Art supports commercial use for ecommerce product displays, marketing materials, and commercial design deliverables. Handle input photos, people, trademarks, and product claims according to the actual project materials. Verifying rights to input assets and product authenticity are separate delivery requirements from the platform's commercial-use support.
Sources and Related Workflows
For model background, see OpenAI's official introduction to ChatGPT Images 2.5, retrieved on 2026-09-26. Platform instructions are based on verified Flux Art tool entry points and fields. This article provides production steps and examples to execute; it does not claim generation tests, pass rates, savings, or customer results.
Find the relevant tools in the Flux Art AI ecommerce workspace. For related tasks, continue with the product production tutorial to learn more about production and review methods.
Related source entry: Related tool or source 1