Generate and edit stills with GPT Image 2.5

GPT Image 2.5 is an image generation and editing model. It accepts a text prompt, and it can also take source images, then return a still. Compared with GPT Image 2, it is built for sharper detail, more natural lighting and texture, stronger subject preservation, and more reliable local edits. In Veo 3 you run it from the generator on this page, with aspect ratio and 1K, 2K, or 4K controls, then collect the file in My Creations.

Use it when the still has to carry a specific subject and a specific layout at the same time. That includes campaign graphics, product scenes, packaging concepts, posters with short lettering, UI mocks, and portrait or product variants that start from a reference. It is also the better GPT Image choice when you expect several refinement passes, because later instructions are more likely to keep earlier decisions in place. It is not a video model; move a finished still to Image to Video if you need motion.

Yes. Open Image to Image, select GPT Image 2.5, upload the source still, and write what must remain plus what should change. Typical edits are background replacement, material or color shifts, wardrobe changes, and short in-image copy. After each result, check the parts you asked to preserve before you stack another instruction. If the subject drifts, shorten the prompt and name those locked details first.

Yes, short lettering is part of what the model is built to handle: posters, labels, ads, and simple UI type. Put the exact words in quotes, state the placement and hierarchy, and inspect every character at the target size. Keep legal copy, prices, dates, and brand-critical typography out of the generation when those strings must be perfect; add them in a design tool after the visual is approved.

Sign in, keep GPT Image 2.5 selected, write one clear still, choose the ratio and resolution, and generate. The button shows the credit estimate for the current settings before you submit. Completed files appear in the preview and in My Creations, so you can download later without repeating the prompt. Paid plans control watermark removal; the generator states that limit when it applies.

This page generates from a prompt. You can choose Auto or a fixed aspect ratio, then 1K, 2K, or 4K. Image to Image adds reference uploads, up to 16 files, when the source already exists. Transparent, opaque, or automatic backgrounds are supported by the model; for high-resolution cutouts, describe an isolated subject with no scenery or shadow. Square 4K is not offered in this generator because that pairing is disabled in the current controls.

Both models generate and edit stills from text and images. GPT Image 2.5 is the newer generation: faster than GPT Image 2, with tighter local edits, more stable reference subjects, and stronger handling of layered briefs that mix layout, lettering, and transparent backgrounds. Keep GPT Image 2 when you want to compare a previous result or stay on a brief that already works there. For new campaign and product work in Veo 3, start with GPT Image 2.5.

The first result appears in the preview on the right. Every completed task is also stored in My Creations, which is the safer place to compare versions, download, and pick a still for Image to Video. Keep a short note with the prompt, ratio, and the one variable you changed so a later pass is a correction, not a new idea. If a task needs support, record its status from My Creations before you write in.