Build a coordinated set of product images with ChatGPT

Imagine a small ceramics business preparing a launch. It needs a square catalogue image, a wide website banner and a portrait social post. All three should show the same blue mug. The practical challenge is to vary the setting and composition while keeping its handle, glaze and proportions recognisable.

On the web, attach a reference image and ask ChatGPT to create or edit it. Availability and limits depend on the plan and workspace. This illustrative workflow follows current documentation; no image-generation test was performed.[2]

Give the reference a specific job

An attached image supplies visual context, but it does not explain your intention. OpenAI distinguishes asking ChatGPT to inspect an image from asking it to create or edit one. Name the task explicitly and, with several attachments, explain what each contributes.[1]

For the ceramics example, begin with a clear photograph of the actual mug. A second image could show the preferred table surface or colour palette. Describe that second attachment as a setting reference so that the request does not accidentally invite a redesign of the mug. Keep the reference set small enough to explain.

OpenAI’s prompting guide recommends assigning roles to references and specifying both the desired changes and the details to preserve. It also warns that repeated edits can alter protected details. If a region must remain identical at the pixel level, its recommendation is to composite the approved edit into the original.[3]

Develop the set from an approved starting point

Approve the product depiction before extending the campaign. If the handle has changed, correcting it now is more useful than creating two more compositions with the same error. Keep the source photograph and approved result distinguishable, so the next request can identify the intended starting image.

In the desktop app, opening a generated image provides Focused view for one image and Canvas view for images from the same chat. Canvas comments can target selected images; Multi-select includes several in the feedback. On the web, continue with a clearly identified reference and a specific editing request.[2]

Request one meaningful revision at a time: move the mug, extend the background or change the crop. Repeat the details that must remain fixed. This follows the documented editing approach and makes it easier to identify which instruction caused an unwanted change.[2]

Review the product and the picture separately

OpenAI’s image-evaluation guidance separates attractive rendering from meeting the brief. It identifies text accuracy, style consistency, correct changes and preservation of other regions as different assessment dimensions. Those distinctions are useful even when a person reviews three images manually rather than building an automated evaluation system.[4]

For this example, compare the handle opening, rim, glaze boundary and base across the set. Then review whether each composition leaves usable space for its intended placement. A pleasing banner that depicts a different handle needs another edit; an accurate mug positioned badly needs a layout change. The corrections serve different purposes.

For a price or slogan, specify the spelling and placement. OpenAI recommends checking dense or production-critical text and finishing in a design tool when necessary. The reserved copy space gives that step somewhere to work.[2]

Save the approved files with names that identify the product, placement and version. Retain the original photograph and the brief alongside them. The next campaign then begins with an identifiable product and a usable creative direction, while the new work can concentrate on the season, setting or offer.

Report an error or suggest a correction