GPT Image by OpenAI · on MossImage
Write a brief or upload a photo, then let GPT Image 2.5 or GPT Image 2 do the work. Box the part of a product shot you want changed, and MossImage keeps every pixel outside the box from your original.
GPT Image is at its best with a brief that reads like art direction. Name the subject and where it sits, describe the setting, then the light, the materials and any words that must appear. “A premium skincare photo” leaves the model to guess; “a frosted glass jar on wet slate, label facing the camera, soft light from the left, empty space at the top for a headline” gives it a plan. Both versions on MossImage accept up to 16 reference images, so a product shot, a logo and a mood picture can go into the same request.
Editing is where MossImage differs from a plain generator. Pick Whole image to change a photo from one written instruction, such as a new background or season. Pick One area to draw boxes and write a note for each one; those box edits run on GPT Image 2.5, and with Keep everything outside the boxes switched on, the untouched parts of your photo stay pixel-identical. Every result is labelled with the version that made it.
Draw a box around a price tag, a background object or a headline and write what should change there. GPT Image 2.5 redraws the boxed areas, then MossImage puts your original pixels back everywhere else. It is the safest way to fix one detail on a listing photo without the product drifting.
Put the exact words in quotation marks and say where they go: “SUMMER SALE” in bold sans-serif across the top third. Short headlines are easier to get right and to proofread. Always check spelling, prices and dates before you publish.
GPT Image 2.5 runs at Standard, Medium, High, Ultra or Max, at 1K, 2K or 4K. Start a new idea at Standard or Medium, then rerun the version you like at a higher tier. The credit cost updates as you change settings, before anything runs.
Straight from the MossImage model catalog. Credits are per image; the editor shows the exact total for your settings before you confirm.
| Version | Can do | References | Sizes | Quality | Aspect ratios | Credits per image |
|---|---|---|---|---|---|---|
| GPT Image 2.5 | Generate · Edit whole image · Box edits · Up to 4 per run | Up to 16 | 1K, 2K, 4K | Standard, Medium, High, Ultra, Max | 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5, 2:1, 1:2, 3:1, 1:3, 21:9, 9:21 | 1K 5–64 · 2K 6–72 · 4K 8–128 |
| GPT Image 2 | Generate · Edit whole image · Up to 4 per run | Up to 16 | 1K, 2K, 4K | Medium, High, Standard | 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5, 2:1, 1:2, 3:1, 1:3, 21:9, 9:21 | 1K 17–65 · 2K 6–258 · 4K 10–516 |
Generate clean product shots on white or in a lifestyle scene, at 1:1 for Amazon or 4:5 for Instagram. Then box any stray detail and fix it without re-rolling the whole photo.
Combine a product reference with a headline, an offer and a layout. Make up to four images in one go with GPT Image 2.5 and keep the strongest one for the campaign.
Describe the face, the expression and the big readable words for a YouTube thumbnail at 16:9, or a story at 9:16. Keep the text short enough to read at phone size.
Start with one subject and a clear composition. With reference images, say what must stay and what may change. The templates below are a good starting point.
Choose the version, shape, size and quality above. The Generate button shows the credits before anything runs.
Check text, faces, hands and product details. To change one part, open the image, choose One area, box it and write what should change; everything outside the box stays as it is.
A vertical product photo of a cobalt-blue glass perfume bottle on pale limestone. The bottle is centered with its label facing the camera, soft window light from the left, gentle reflections. Keep the top third empty for a headline. No text.
Material, label direction and headline space come before style words.
Use this promptA square sale poster for running shoes. A red running shoe floats at an angle over a warm orange gradient. Headline “48-HOUR SALE” in heavy white sans-serif at the top, “UP TO 40% OFF” smaller below it. Clean, bold, high contrast.
Quote every word that must appear and say how big it should be.
Use this promptOpen any example for its full prompt and the exact version that made it. A result from one version does not prove what another version will do.
GPT Image 2.5 and GPT Image 2, both by OpenAI. They are separate choices in the model menu with their own sizes, quality tiers and credits. Each image you make is labelled with the exact version.
Open the image in the editor, choose One area, draw a box and write what should change there. You can add several boxes, each with its own note. Box edits run on GPT Image 2.5, and Keep everything outside the boxes restores your original pixels outside them.
Start with GPT Image 2.5. It is cheaper at most settings, offers five quality tiers and handles box edits. GPT Image 2 is useful for comparing results on the same brief. Run both on one prompt and keep the version you prefer.
Yes, when you quote the exact wording and say where it goes. It is not guaranteed to be perfect, so check every word, number and price before you use the image. Shorter text is easier to get right.
It depends on the version, resolution and quality tier: GPT Image 2.5 starts at 5 credits for a 1K image at Standard. The table above lists every setting, and the editor shows the exact total before you confirm. Failed jobs are refunded automatically.
Only you. Uploads and generated images are private to your account and are not indexed by search engines. Your images are kept in History, where only you can open them.