GPT Image 2.5
OpenAI's newest image model, the one behind ChatGPT Images 2.5. Generate and edit with Flare or Sunburst right here, without a ChatGPT account. 8 credits an image on the default tier, and new accounts start with free credits.
Flare is selected. Pick Sunburst in the model menu when detail matters more than speed, or GPT Image 2 to compare the same prompt.
GPT Image 2.5 in your browser
OpenAI released GPT Image 2.5 on September 8, 2026, as two models. Flare is built for speed and volume. Sunburst takes longer and holds edits more tightly. Both run in the generator above, with up to 16 reference images, 15 aspect ratios plus Auto, and 1K to 4K output. Every example on this page was generated with GPT Image 2.5, and each caption names the model.
GPT Image 2.5 Flare vs Sunburst on one prompt
Same prompt, same settings (2K, high quality), sent to both models at the same time on this site.
Sunburst spends its extra time on texture
Both models got the chalkboard right. Flare came back in 41 seconds and stayed close to the prompt. Sunburst took 74: the crusts carry flour, the condensation on the glass is heavier and the backlight is harsher. It also added signage nobody asked for, a BAKERY sign and a street name, so read the whole frame before you ship a Sunburst image.
A corner bakery window at seven in the morning, photographed from the pavement. A chalkboard easel inside the glass reads 'FRESH RYE 3.50' on the first line and 'OPEN UNTIL SOLD OUT' underneath in hand-lettered chalk. A tray of dark crusty loaves sits on a wooden counter, steam rising and catching low golden sunlight. The glass is slightly fogged at the edges with faint reflections of the street. Photorealistic, natural colour, no people.
Text that stays readable at full size
Posters, packaging, shop signage, slide graphics. This poster came out of GPT Image 2.5 Flare on the first run: crisp letterforms, every word spelled right, the thin rule exactly where the prompt put it. Pages crowded with tiny labels still trip it up, so keep small text short.
A minimalist letterpress concert poster: a 'NORTHERN LIGHTS' headline over 'Chamber Ensemble - Hall Two - 8 PM', a thin rule between them, and 'TICKETS AT THE DOOR' along the bottom edge. Deep indigo, cream ink, subtle paper grain.
GPT Image 2.5 editing in one sentence
Say what to change in plain words: swap an object, recolour a surface, move the subject somewhere new. OpenAI says 2.5 follows edit instructions more reliably across several rounds, so you can keep working on one image instead of starting over.
Change one thing, keep the scene
This edit ran on GPT Image 2.5 Sunburst, the model OpenAI positions for tighter control across edits. The notebook became an open sketchbook and the mug went matte black. The desk, wall, plant and glasses came through, though a few objects moved slightly on the desktop, so check the layout when placement matters.
Replace the closed notebook with an open sketchbook showing a pencil drawing of a bicycle, and make the mug matte black.
Keep the subject, change the world around it
Better preservation of subjects from reference photos is one of the changes OpenAI lists for 2.5. Here GPT Image 2.5 Flare moved the knitted fox onto a forest log and kept the yarn texture, the crooked ear and the button eyes.
Put this exact plush fox on a mossy log in a misty pine forest at dawn.
Product shots without the shoot
Listing images, launch graphics, ad variants. OpenAI built GPT Image 2.5 Sunburst for polished product images like this one, and it is slower than Flare for that reason. Name the product, the surface and the light. When you need forty variants, run them on Flare and save Sunburst for the final frame.
An e-commerce photograph of a frosted glass skincare bottle with a matte sage-green cap on a travertine block, hard afternoon sun, a dried eucalyptus sprig beside it, the label reading FIELD & FERN.
How to use GPT Image 2.5 online
Pick a mode
Text to Image opens with GPT Image 2.5 Flare selected. Choose Sunburst in the model menu for slower, more careful results, or switch to Image to Image to edit a photo you upload.
Describe it properly
Name the subject, the light, any text that has to appear, and the layout. Set the aspect ratio from 16 options, plus resolution and quality.
Generate, then adjust
At high quality and 2K, Flare returned our test images in 41 to 48 seconds and Sunburst in 68 to 74; the default 1K tier is quicker. Both cost 8 credits on the default tier. Change a few words or feed the result back in as a reference, then download the one you keep.
GPT Image 2.5 Flare vs Sunburst vs GPT Image 2
The four OpenAI models you can run on this page. Credits are for the default tier: medium quality, 1K.
| Feature | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst | GPT Image 2 | GPT Image 1.5 |
|---|---|---|---|---|
| Text-to-image generation | Yes | Yes | Yes | Yes |
| Image-to-image editing | Yes, up to 16 refs | Yes, up to 16 refs | Yes, up to 5 refs | Yes, up to 16 refs |
| Aspect ratios | 16 | 16 | 16 | 4 |
| Output resolution | 1K–4K | 1K–4K | 1K–4K | Fixed size |
| Images per generation | 8 | 8 | 8 | 1 |
| Credits / image | 8 | 8 | 8 | 7 |
| Best for | Everyday work and volume. OpenAI cites about 50% lower latency than GPT Image 2 | Final product and campaign images. Slower, with tighter control over edits | Prompts you already tuned for GPT Image 2 | Simple drafts at a lower price |
Explore more AI image models
GPT Image 2
8 credits / image
The default here before 2.5 arrived. Still worth a run when a prompt you tuned for it looks better on the older model.
Try nowSeedream 5.0 Pro
5 credits / image
ByteDance's Seedream 5.0 Pro — 2K output and strong bilingual English and Chinese text.
Try nowNano Banana 2 Lite
4 credits / image
Google's cheapest Gemini image tier, for fast drafts when you only need the idea on screen.
Try nowGPT Image 2.5 FAQ
Try GPT Image 2.5 Flare and Sunburst free
Open the full editor, pick a ratio and quality, and generate from a text prompt or up to 16 reference images.