Image model · by OpenAI
GPT Image 2.5 Flare AI image generator
GPT Image 2.5 Flare, the smaller of OpenAI’s two GPT Image 2.5 models, is DreamBench’s default image model. It’s also the model DreamBench’s video engine reaches for first when it draws the character, prop and location sheets your videos are built from.
Available on every plan · Released September 2026
- 16 credits per image
- Up to 10 images per run
- 6 aspect ratios
- Character, prop and location sheets
- On every plan
Start with a prompt
Try it now
Edit the idea or write your own. Press Create with GPT Image 2.5 and the model is already picked for you.
With any DreamBench plan
Tools that run on GPT Image 2.5
Each tool takes one photo and returns one result made by GPT Image 2.5, ready to download full size or to turn into a video.
Strengths
What GPT Image 2.5 is good at
- 01
The look of DreamBench’s own sheets
DreamBench’s video engine reaches for GPT Image 2.5 Flare first when it draws the sheets a video is built from, so the images you make with it here share their look.
- 02
Sheets your videos can use
Pick a character, prop or location sheet, or a location schematic, and DreamBench builds it with the templates it uses for a video’s own sheets. Add it to a video and it’s used as it is, not redrawn.
- 03
Pictures as well as words
Flare takes images as well as text, so a brief can come with reference pictures of the characters, products or places to include. Mention each one by its @name and DreamBench tells the model which picture is which.
At a glance
GPT Image 2.5 on DreamBench
credits per image
images per run
aspect ratios
1:1, 4:3, 3:4, 16:9, 9:16 and 21:9
things to make
Freeform images, character sheets, prop sheets, location sheets and location schematics
The model
What is GPT Image 2.5?
- Made by OpenAI
- Released September 2026
- Image model
GPT Image 2.5 is a pair of image models from OpenAI, Sunburst and Flare, released on September 8, 2026, for making and editing images. DreamBench runs GPT Image 2.5 Flare, the smaller model: OpenAI optimized it for speed and aims it at high-quality images for everyday use, and it can work from a text brief, reference images or both.
OpenAI describes Flare’s image quality as comparable to GPT Image 2. Sunburst, the base model, is the one OpenAI optimized for quality, and DreamBench doesn’t offer it. OpenAI also notes that a complex prompt can take as long as 2 minutes to process.
More from OpenAI: GPT Image 2.5 Flare on OpenAI’s developer platform
The details
GPT Image 2.5 specs and credits
What you get when you make images with it on DreamBench.
| Made by | OpenAI |
|---|---|
| Credits | 16 per image |
| Images per run | 1 to 10 |
| Aspect ratios | 1:1, 4:3, 3:4, 16:9, 9:16 and 21:9 |
| What you can make | Freeform images, character sheets, prop sheets, location sheets and location schematics |
| Reference images | Up to 8 per run |
| Plans | Every plan |
Good to know
- Picking GPT Image 2.5 on DreamBench always means GPT Image 2.5 Flare, OpenAI’s model gpt-image-2.5-flare, and every image in the run is made on it: DreamBench doesn’t swap in another model. GPT Image 2.5 Sunburst isn’t offered, so what OpenAI says about Sunburst doesn’t apply here.
- OpenAI charges for every picture the model reads, so each reference image raises the price of every image in the run. The composer shows the exact cost before you generate.
- Ask for more than one freeform image and the first is your brief exactly as written. Each of the others takes a direction of its own, such as a closer view or another palette, and keeps what your brief spells out.
- Through OpenAI’s own API, GPT Image 2.5 Flare can also edit an existing image or make one with a transparent background. On DreamBench, it makes new images from your brief and any reference images you add.
Estimate your credits
A GPT Image 2.5 run, as the composer prices it.
64credits
16 credits per image. Starter’s 1,100 monthly credits make about 68 images. You always see the exact cost before you generate.
Side by side
GPT Image 2.5 compared
GPT Image 2.5 beside the models people weigh it against, as each runs on DreamBench. Every cell comes from the same tables as the spec sheet above.
| Model | Images per run | Aspect ratios | What it makes | Reference images | Plans |
|---|---|---|---|---|---|
| GPT Image 2.5OpenAI | 1 to 10 per run | 1:1, 4:3, 3:4, 16:9, 9:16, 21:9 | Freeform images, character sheets, prop sheets, location sheets and location schematics | Up to 8 | Every plan |
| GPT Image 2OpenAI | 1 to 10 per run | 1:1, 4:3, 3:4, 16:9, 9:16, 21:9 | Freeform images, character sheets, prop sheets, location sheets and location schematics | Up to 8 | Every plan |
| Nano Banana 2Google DeepMind | 1 to 10 per run | 1:1, 4:3, 3:4, 16:9, 9:16, 21:9 | Freeform images, character sheets, prop sheets, location sheets and location schematics | Up to 8 | Every plan |
| Seedream 5.0ByteDance Seed | 1 to 10 per run | 1:1, 4:3, 3:4, 16:9, 9:16, 21:9 | Freeform images, character sheets, prop sheets, location sheets and location schematics | Up to 8 | Every plan |
How it works
How to make images with GPT Image 2.5
Step 1
Choose GPT Image 2.5
Pick GPT Image 2.5 in the composer’s image mode, or press Create with GPT Image 2.5 on this page.
Step 2
Describe the image
Write what you want to see, or make a character, prop or location sheet for a video you’re planning.
Step 3
Generate and reuse
Make up to 10 versions at once. Every image lands in your library, ready to use as a reference in your videos.
Prompt ideas
Prompt ideas for GPT Image 2.5
Start from one of these and change it however you like.
A character to cast
A full-length 3:4 portrait of an invented character for a short film: a lighthouse keeper in her sixties with cropped silver hair, a navy fisherman’s sweater, yellow oilskin pants and a brass lantern in one hand. She stands on wet rocks at dusk and looks at the camera with a wry half-smile. Soft, even light, so her face, clothes and lantern read clearly.
A place to set a scene
A wide 21:9 view of a floating market at dawn. Wooden boats piled with mangoes, chillies and marigolds crowd a canal between stilt houses, mist lifts off the water and the first sun catches the tin roofs. Keep the foreground clear and the canal open into the distance, so the place reads as a setting a story could move through.
A teapot on the table
A square 1:1 still life of a handmade ceramic teapot in a speckled sea-green glaze on a scrubbed oak table, steam curling from the spout. A folded linen napkin, 2 small cups and a sprig of rosemary sit beside it. Late-morning window light from the left, soft shadows, a shallow depth of field and a calm, warm mood.
A magazine cover
A 3:4 cover for a travel magazine. A woman in a red raincoat stands on a windswept harbor wall at dawn, fishing boats behind her. The masthead across the top reads “WANDER” in tall white serif capitals, with “Issue 12” in small type beside it, and a cover line at the bottom reads “48 hours on the Atlantic coast”.
A box on the shelf
A 1:1 packaging shot of a matte kraft-paper tea box on a pale wooden shelf, lit by soft daylight from the right. A line drawing of a mountain fills the front, with “Highland Breakfast” in dark green type above it and “20 tea bags” in small capitals below. A ceramic mug and a few loose tea leaves sit beside the box.
Still have questions?
GPT Image 2.5 FAQ
What is GPT Image 2.5?
GPT Image 2.5 is a pair of image models from OpenAI, Sunburst and Flare, released in September 2026 for making and editing images. DreamBench runs GPT Image 2.5 Flare, the smaller model, which OpenAI optimized for speed and aims at high-quality images for everyday use.
Which GPT Image 2.5 model does DreamBench use?
DreamBench uses GPT Image 2.5 Flare, OpenAI’s gpt-image-2.5-flare, for every image in a run. It’s DreamBench’s default image model and the one its video engine reaches for first when it draws a video’s character, prop and location sheets. OpenAI describes Flare’s image quality as comparable to GPT Image 2, and GPT Image 2.5 Sunburst isn’t offered.
What is the difference between GPT Image 2.5 Flare and Sunburst?
GPT Image 2.5 Flare and Sunburst are OpenAI’s two GPT Image 2.5 models, released on September 8, 2026. Flare is the smaller model, optimized for speed, with image quality OpenAI calls comparable to GPT Image 2; Sunburst is the base model, optimized for quality, with longer generation times. DreamBench runs Flare for every GPT Image 2.5 image, so what OpenAI says about Sunburst doesn’t apply here.
What can I make with GPT Image 2.5 on DreamBench?
With GPT Image 2.5 on DreamBench you can make freeform images, plus character, prop and location sheets and location schematics. It’s also the model DreamBench’s video engine reaches for first when it draws a video’s sheets, and a sheet you make here is used as it is when you add it to a video.
Can I use reference images with GPT Image 2.5?
Yes. Attach reference images in the composer and mention each by its @name in your brief, and DreamBench tells GPT Image 2.5 which picture each name stands for. Each reference raises the price of every image in the run.
How much does GPT Image 2.5 cost on DreamBench?
GPT Image 2.5 uses credits for each image, so a run of several versions costs more than a single 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.
Does GPT Image 2.5 cost anything to use?
Yes. On DreamBench, each GPT Image 2.5 image uses credits, and credits come with a paid plan, with GPT Image 2.5 on every plan. Each reference image raises the price of every image in the run, and the composer shows the exact cost before you generate.
What should a GPT Image 2.5 prompt say?
OpenAI’s guide says to set out the scene, the subject, its details and your constraints, in that order for a complex brief, to put any exact lettering in quotation marks and say where it goes, and to state in words what to leave out. On DreamBench, mention each reference image by its @name, and allow time: OpenAI notes a complex prompt can take as long as 2 minutes.
Can GPT Image 2.5 put text in images?
Yes. Put the exact words in double quotation marks and describe where they sit, as OpenAI’s prompting guide advises, and DreamBench keeps them character for character in every version of the run. OpenAI notes its GPT Image models can still misplace or blur words, so check the lettering before you use the image.
Can I edit an existing image or get a transparent background with GPT Image 2.5 on DreamBench?
No. On DreamBench, GPT Image 2.5 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.
Should I use GPT Image 2.5 or Nano Banana 2 on DreamBench?
Both are in DreamBench’s image mode, so run the same brief on each and keep the version you prefer. GPT Image 2.5 Flare is DreamBench’s default image model and takes up to 8 reference images, each raising the price; the Nano Banana 2 page, linked below, states its own facts.
Does DreamBench have a GPT Image 2.5 API?
No. DreamBench offers GPT Image 2.5 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.
Do GPT Image 2.5 images have a watermark on DreamBench?
No. Images you make on DreamBench carry no DreamBench watermark, on any plan, and come to you as GPT Image 2.5 Flare generated them.
Can I use GPT Image 2.5 images commercially?
Yes. DreamBench’s Terms of Use allow personal and commercial use, including ads and client work. You need the rights to anything you upload.
Which DreamBench plans include GPT Image 2.5?
GPT Image 2.5 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.
How many images can GPT Image 2.5 make in one run?
From 1 to 10 in a single run, each at 16 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.
How do I make images with GPT Image 2.5 on DreamBench?
Press Create with GPT Image 2.5, 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.
Keep exploring
Other models to try
OpenAI
GPT Image 2
Put words inside the picture, from a poster headline to a speech bubble, or draw sheets for your videos.
61 credits per image
Google DeepMind
Nano Banana 2
Make posters, cards and diagrams with lettering, or design the characters and places for your next video.
6 credits per image
ByteDance Seed
Seedream 5.0
Make infographics, posters and photographic images, with text in many languages.
15 credits per image
Where the facts come from
Sources
- GPT Image 2.5 Flare on OpenAI’s developer platform
- OpenAI’s GPT Image 2.5 prompting guide
- OpenAI’s image generation guide
GPT Image 2.5 is developed by OpenAI. DreamBench is an independent platform that offers GPT Image 2.5 alongside other models, and isn’t affiliated with or endorsed by OpenAI. The facts on this page were last checked on . GPT Image 2.5 and OpenAI are trademarks of their owners, used here only to name the model and its developer.
Make images
Create withGPT Image 2.5
Describe the image you want, or build the characters and places for your next video.