GPT Image 2 AI image generator

GPT Image 2 is an image model from OpenAI, released in April 2026 for making and editing images. It can render text inside an image, and it reads every image you give it at high fidelity.

Yes. OpenAI lists text rendering among GPT Image 2’s abilities, so a headline, sign or speech bubble can be part of the picture, though it notes its GPT Image models can still misplace or blur words. On DreamBench, put the exact words in double quotation marks and they stay character for character in every version of the run; keep them short, ask for several versions to pick the one that reads right, and check the lettering before you publish.

Both developers list text among their models’ abilities, and both note that lettering can still come out wrong. The same goes for Seedream 5.0 Pro, which ByteDance built for text-rich layouts. All three are in DreamBench’s image mode, so run the same brief on each and keep the version that reads best.

With GPT Image 2 on DreamBench you can make freeform images, plus character, prop and location sheets and location schematics drawn to the same templates DreamBench uses inside a video run. Everything you make is saved to your library for your next video.

GPT Image 2.5 is OpenAI’s follow-up to GPT Image 2, released in September 2026 as two models, Flare and Sunburst. DreamBench runs GPT Image 2 and GPT Image 2.5 Flare, whose image quality OpenAI describes as comparable to GPT Image 2, so you can run the same brief on each and keep the version you prefer.

GPT Image 2 uses credits for each image, so a run of several versions costs more than one image, and each reference image raises the price of every image. The table on this page shows the current rate, and the composer shows the exact cost before you generate.

Yes. On DreamBench, each GPT Image 2 image uses credits, which come with a paid plan, and GPT Image 2 is on every plan. The composer shows the exact cost before you generate.

Write the brief as instructions in plain sentences, with the subject first and exact lettering in quotation marks. Say what to leave out in words, as OpenAI’s guide advises: there is no negative-prompt field. On DreamBench, mention each reference image by its @name.

No. On DreamBench, GPT Image 2 makes new images from your brief and reference images, with full backgrounds. Editing an uploaded image and transparent output are features of OpenAI’s own API.

No. DreamBench offers GPT Image 2 through its composer, in the browser on a phone, tablet or computer, with nothing to install. Developers can use OpenAI’s own API, linked under Sources, where the prices and limits are OpenAI’s, not DreamBench’s.

No. Images you make on DreamBench carry no DreamBench watermark, on any plan, and come to you as GPT Image 2 generated them.

Yes. DreamBench’s Terms of Use allow personal and commercial use, including ads and client work. You need the rights to anything you upload.

GPT Image 2 is on every DreamBench plan, from Starter at $9.99 a month. Every plan makes freeform images, character sheets, prop sheets, location sheets and location schematics with it, up to 10 images in one run.

From 1 to 10 in a single run, each at 61 credits, in any of 6 aspect ratios (1:1, 4:3, 3:4, 16:9, 9:16 and 21:9). You see the exact cost before you generate.

Press Create with GPT Image 2, or pick it in the composer’s image mode, describe the image, choose how many and the aspect ratio, and generate. Add up to 8 reference images to steer the look.