OpenAI's model behind ChatGPT Images 2.5, for posters, labels, diagrams and photos. Up to 4K, with up to 14 reference images to edit or combine.
Four ideas across photo, painting and illustration. Click one to load it into the generator, then change the details.
IllustrationText to image · 1:1 · 1K
Prompt: A cozy reading nook with a green velvet armchair, a knitted throw and a tall wooden bookshelf, soft afternoon sunlight through a window, plants, photorealistic interior, no text
IllustrationText to image · 1:1 · 1K
Prompt: A watercolor painting of a white lighthouse on a rocky cliff at sunset, pink wildflowers in the foreground, loose brush strokes, warm sky, no text
IllustrationText to image · 1:1 · 1K
Prompt: A surprised young man in a purple hoodie and a yellow beanie holding a giant slice of pepperoni pizza with stretchy cheese, bright yellow and blue background, studio photo, no text
IllustrationText to image · 1:1 · 1K
Prompt: An isometric illustration of a small bright office with four people working at desks, plants, a bookshelf and big windows, soft colors, no text
GPT Image 2.5 is OpenAI's image model behind ChatGPT Images 2.5, released in September 2026. It makes new images and edits photos you upload, and it is a common pick for posters, labels and diagrams, where the words have to read correctly.
On Imgveo it renders at 1K, 2K or 4K in 8 aspect ratios, costs 10 credits per image at 1K, and takes up to 14 reference images in the generator above. Failed images are refunded.
OpenAI offers the model in two versions. Imgveo runs Flare, the fast default, because quick turns suit trying a prompt, reading the result and trying again.
What you can generate right now. Every number comes from the generator's own settings.
OpenAI ships GPT Image 2.5 as two models with different trade-offs. Most write-ups mix them up; here is the difference.
OpenAI's default. It favors speed, so you can iterate on a prompt in quick rounds. Every GPT Image 2.5 image on Imgveo is made with Flare.
Slower, and tuned by OpenAI for precise edits. Imgveo doesn't offer it; if you need heavier editing, Nano Banana Pro is the alternative here.
All three run on Imgveo and all three reach 4K. GPT Image 2.5 and 2 cost the same; Nano Banana Pro costs more per image.
The default for posters, labels, diagrams and edits from several reference photos.
Same price, more aspect ratios, including 4:5 and 9:21. Keep it for prompts and styles that already work on it, or when you need one of those extra formats.
Google's heavier model, for text in many languages and the most detailed images.
| GPT Image 2.5 | GPT Image 2 | Nano Banana Pro | |
|---|---|---|---|
| Maker | OpenAI | OpenAI | |
| Credits per image, 1K | 10 | 10 | 28 |
| Credits per image, 2K | 15 | 15 | 28 |
| Credits per image, 4K | 25 | 25 | 36 |
| Aspect ratios | 8 | 11 | 10 |
Prices without reference images. Adding them costs a little more; see the price table below.
No setup and no ChatGPT subscription: the generator at the top of this page runs it.
Describe a new image, or add up to 14 reference images: photos to edit, products to combine, or a rough sketch.
Put the exact words in quotes and say where each goes: headline at the top, labels on the parts, small print at the bottom.
The button shows the exact credits. Draft at 1K, then render the one you keep at 2K or 4K.
Headlines, labels and lettering are where GPT Image 2.5 earns its place. Four layouts, with the prompt behind each.
IllustrationPrompt
A retro 1970s magazine cover titled "FUTURE KITCHENS" with the line "The Magazine of Tomorrow's Home Today", a photo of a colorful orange kitchen, cover lines "Color Comes to the Kitchen!" and "Design Ideas for the 1970s Home", "OCT 1974" and "75¢" in the corner
IllustrationPrompt
A labeled diagram titled "Parts of a Bicycle": a blue bicycle on a white background with arrows to Saddle, Handlebars, Frame, Front Fork, Chain, Pedals, Brake and Tire, clean sans-serif labels in rounded boxes
IllustrationPrompt
A wedding invitation card with elegant script lettering that reads "Anna & Leo", pink roses and white flowers around the border, a thin gold frame, cream paper
IllustrationPrompt
A cardboard shipping box on a plain light gray background with the brand name "NORTHWIND" and a mountain logo printed in navy on the side, product mockup
Draw the layout by hand, photograph it and upload it as a reference. Then describe the finished look: "Turn this sketch into a clean flat illustration, keep the layout." The boxes and arrows become the composition.
Headlines, labels and names go in quotation marks, exactly as they should appear.
"Magazine cover", "labeled diagram", "centered invitation": a layout word sets the structure.
"Bold 1970s lettering", "elegant script", "clean sans-serif labels" all steer the look.
Text comes out clean most of the time, not every time. Check names, numbers and spelling.
Layout, subject, text, style, then format. Five parts, in that order, cover posters, packaging and photos alike.
What kind of image: cover, poster, diagram, product shot, portrait.
The main thing in it, with two or three concrete details.
Every word in quotes, with where it goes.
Era, medium and color: 1970s print, watercolor, flat vector, studio photo.
Aspect ratio and resolution, set in the generator: 3:4 for covers, 1:1 for products.
Replace the words in brackets.
Text to image · 3:4 · 2K
A [era or style] [magazine or book] cover titled "[TITLE]" with the line "[subtitle]", [main image], cover lines "[line 1]" and "[line 2]", [color palette]
Text to image · 1:1 · 2K
A labeled diagram titled "[TITLE]": [object] on a white background with arrows to [part 1], [part 2], [part 3] and [part 4], clean sans-serif labels
Text to image · 1:1 · 2K
A [box, bag or bottle] on [background] with the brand name "[BRAND]" and [logo idea] printed in [color], [lighting], product mockup
Image edit · 4:3 · 2K
Turn this sketch into a [style] image. Keep the layout: [what is where]. Add [colors and details], and set the text "[TEXT]" where the sketch shows it
4K costs 25 credits; get the layout right at 1K for 10 first.
"A title about cooking" gets you a random title. Write the title.
Ten labels on a small image blur together. Fewer, larger words read better.
You pay per image, by resolution, the same as GPT Image 2. Every number below comes from the same code that prices the generator.
| Resolution | Text to image | With reference images | Nano Banana Pro |
|---|---|---|---|
| 1K | 10 | 12 | 28 |
| 2K | 15 | 18 | 28 |
| 4K | 25 | 35 | 36 |
One image costs 10 credits at 1K, 15 at 2K and 25 at 4K. Four images at once cost four times as much.
The Starter plan's 1,500 monthly credits cover about 150 GPT Image 2.5 images at 1K, or 100 at 2K.
If an image fails, its credits go back to your balance automatically. You only pay for the images you receive.
Jobs where the words and the layout matter as much as the picture.
Magazine, book and event covers with a headline set the way you wrote it.
Labeled parts, arrows and steps for lessons, manuals and blog posts.
Names and dates in script or serif lettering, framed with flowers or patterns.
Your brand name on boxes, bags and labels before anything is printed.
A product, a headline and a call to action in one square image.
Isometric scenes, watercolor and flat vector art for sites and slides.
Open Imgveo Studio to run this and every other AI video and image model from a single workspace, with one credit balance.
Compare GPT Image 2.5 with the models people weigh it against.
Compare: GPT Image 2.5 vs Nano Banana Pro
Write your prompt with the words in quotes, or upload a sketch or photos, and generate at up to 4K. Failed images are refunded.
ChatGPT and GPT are trademarks of OpenAI. Imgveo is an independent platform and is not affiliated with OpenAI.