Describe the layout and the wording, or upload an image with one line about what to change, and get a finished graphic back.
Four capabilities OpenAI built into this release, described in terms of what comes back on the canvas.
Diagrams, infographics, charts, posters and comics arrive with their sections already arranged: the callout sits beside the part it points to, the poster keeps a clear headline hierarchy, the comic keeps its panels apart. You are assembling one image, not stitching a render into a page layout afterwards.
Headlines, chart labels, button captions and speech balloons come out sharper and easier to read than earlier attempts at GPT Image 2 text in images, and multiple languages are supported inside the same frame.
A brief carrying a headline, a date, a layout requirement and a style note tends to come back placed as written. Fewer rounds of the same request, reworded.
Upload a product photo, an earlier poster or a screenshot and say in plain words what should change. High fidelity image inputs also work as a style reference, so the GPT Image 2 photo editor side and the generator side are one model rather than two tools.
How a brief becomes a published graphic on VideoAI, from the description through to the aspect ratio you download.
Write the wording that must appear on the image exactly as you want to see it, headline, section labels, captions and dates, then say how the panels or columns should sit. A line such as a three panel diagram with each panel labelled gives the model both the copy and the structure it needs.
Already have the product photo, the last version of the poster or a screenshot? Upload it and say what should change in one sentence, such as swapping the background or rewriting the line on the packaging. The same model handles both jobs, so nothing moves to another tool.
Pick the ratio for where the graphic will live: 4:3 for a newsletter body, 16:9 for a thumbnail or a wide post, 1:1 for a square feed, 9:16 for phone mockups and Stories. Deciding the destination before you generate saves a recrop later, and results come back at high resolution, ready to download.
Browse additional AI models for image and video creation.
Describe the layout, spell out the words, and take away a graphic you can put in front of readers.
Start NowIt is an image generation and editing model from OpenAI, released in April 2026. The openai GPT Image 2 release focuses on placing what you describe where you asked for it, rendering readable wording inside the picture, and holding a structured layout together across diagrams, posters and comic panels.
Yes, those are two names for one model. You may also see it written in lowercase as GPT Image 2. This page uses that spelling throughout so there is no confusion about which release you are picking.
Write the exact wording you want printed on the graphic together with the layout you have in mind, choose your aspect ratio, and generate. Treating the prompt like a brief to a designer, copy first and structure second, gives the closest result on the first pass.
It can. Upload the file, then describe the change in plain language: a new background for a shop listing, a different season behind the product, one reworded line on the packaging. Generation and editing both run through this GPT Image 2 photo editor, so nothing needs uploading again elsewhere.
Text rendering improved in this release and supports multiple languages, so headlines, chart labels, interface captions and comic dialogue tend to come out clearly legible. Long or unusual strings still deserve a look before you publish, which is why spelling them out in the brief helps.
Structured visuals are a strength here. OpenAI points to diagrams, infographics, charts, posters and comics as work this GPT Image 2 image generator handles, which covers newsletter graphics, explainer visuals and carousel cards.
Output arrives at high resolution, and on VideoAI you can select 1:1, 4:3, 16:9 or 9:16 before generating. Choosing the ratio that matches your publishing slot keeps the layout intact instead of forcing a crop through the middle of a label.
New users get 10 credits at no cost on VideoAI, and that starter balance covers this model directly. Signing in is required before you generate. Once the credits run out you can top up with a credit pack or move to a subscription plan.
Nothing carries a watermark, including images made with the starter credits, and you can download them straight away in common image formats. Commercial use is permitted, so newsletter graphics, shop listings and client deliverables are all fine.
