New: GPT Image 2.5 is live. Sharper images, faster generation and precise editing in Workspace.Try Now

GPT Image 2.5

GPT Image 2.5 is OpenAI's newest image model, released September 8, 2026. Generate and edit images online with GPT Image 2.5 Flare for speed or GPT Image 2.5 Sunburst for precision: sharper detail, faithful reference photos, targeted edits that leave the rest untouched, and 1K / 2K / 4K output with up to 16 reference images.

Reference Fidelity

GPT Image 2.5 keeps faces, products, and places recognizable when you move them into a new scene, style, or outfit. Upload up to 16 reference images and the distinctive features carry through instead of drifting.

Edit Only What You Ask

Change a wall color, swap a pillow, or rewrite a headline while GPT Image 2.5 leaves the composition, lighting, and everything else exactly where it was. Earlier edits survive later ones across multi-turn refinement.

Natural Light & Rich Texture

Compared with GPT Image 2, GPT Image 2.5 renders softer, more believable lighting and richer material detail: fabric, glass, skin, stone, and metal all read as photographed rather than painted.

Flare for Speed, Sunburst for Polish

GPT Image 2.5 Flare cuts latency by up to 50% versus GPT Image 2 and is the default for everyday work. GPT Image 2.5 Sunburst spends more time per image for tighter control on production-ready assets. Same price, one-click switch.

Complex Briefs, One Pass

Layered prompts with multiple subjects, spatial relationships, in-image text, brand styling, and transparent backgrounds come out coherent. GPT Image 2.5 follows long, multi-clause instructions without dropping requirements.

1K, 2K and 4K in 13 Ratios

Pick 1K for fast drafts, 2K for social and web, 4K for print and campaign hero images. GPT Image 2.5 supports 13 aspect ratios from 21:9 cinematic to 9:16 vertical, plus auto framing that follows your reference.

Same Person, New Scene, Nothing Lost

This is the headline improvement in GPT Image 2.5. Upload a reference photo and ask for a new outfit, setting, or era, and the model preserves the identity that matters: face shape, expression, hair, and the small details that make a subject recognizable. The same holds for products and places. On the Arena image-edit leaderboard GPT Image 2.5 Sunburst and GPT Image 2.5 Flare hold the top two spots, ahead of GPT Image 2, Seedream 5.0 Pro, and Nano Banana Pro, and reference fidelity is a big reason why. It makes avatar tools, character systems, and product-scene generation dependable instead of a lottery.

Prompt

Take the boy from the reference school photo and dress him in a white tuxedo with a black bow tie for a formal portrait. Keep his face, expression, and skin tone exactly the same; only change the outfit and lighting.

Reference → formal portrait

Reference → formal portrait

One reference, four consistent variations

One reference, four consistent variations

Targeted Edits That Stay Targeted

Ask GPT Image 2.5 to change one thing and it changes one thing. Recolor a wall and the bedding, lamps, rug, shadows, and morning light stay put. Swap a cushion and the sofa, throw, plants, and camera angle do not shift. This precision also compounds across turns: GPT Image 2.5 carries earlier adjustments forward, so a five-step refinement ends where you intended instead of quietly drifting. For product retouching, real-estate staging, and campaign iteration this is the difference between a usable tool and a novelty.

Prompt

Change only the pale gray wall color to a muted warm sage green. Keep the bed, bedding, lamps, curtains, framed print, rug, book, glasses, flooring, shadows, morning light, and camera angle unchanged. Preserve realistic wall texture and natural color interaction with the existing daylight.

Wall recolored, everything else untouched

Wall recolored, everything else untouched

Cushion swap with stable composition

Cushion swap with stable composition

Posters, Text and Layout in One Shot

GPT Image 2.5 inherits the in-image text rendering that made GPT Image 2 popular and adds better layout discipline. Headlines land where a designer would put them, dates and venues sit in a proper footer, and a four-poster set keeps a consistent palette and typographic voice across every frame. Dense small type is still the hardest case for any image model, so keep body copy short, but for event posters, merch graphics, social tiles, and packaging concepts GPT Image 2.5 gets you to a presentable draft in a single generation.

Prompt

Elegant editorial poster for 'Wellness Fest': oversized serif headline in deep green and terracotta, pressed peony illustration, a small card in the corner reading 'March 23 · Seaside Arts Hall', muted sage background, generous whitespace.

Event poster with legible headline and footer

Event poster with legible headline and footer

Four-frame travel poster set, consistent style

Four-frame travel poster set, consistent style

Photoreal Light, Skin and Material

The most visible upgrade from GPT Image 2 to GPT Image 2.5 is how light behaves. Golden-hour rim light wraps a glass perfume bottle correctly, jasmine petals cast soft shadows on rough stone, and a studio portrait shows natural skin texture and fabric drape instead of the plastic smoothness older models produced. GPT Image 2.5 is built on OpenAI's autoregressive image architecture rather than diffusion, which is why it follows instructions so literally while still delivering this level of realism. Product teams, photographers, and e-commerce sellers get shots that look captured, not generated.

Prompt

Luxury perfume bottle labeled 'Aurelia' standing on a weathered stone ledge above the Mediterranean at golden hour, white jasmine flowers around the base, soft warm rim light, shallow depth of field, editorial product photography.

Product shot with realistic glass and light

Product shot with realistic glass and light

Studio portrait with natural skin and fabric

Studio portrait with natural skin and fabric

Use cases

Product Scenes from One Reference

Product Scenes from One Reference

Upload a single packshot and let GPT Image 2.5 place the product in a bathroom shelf, a lifestyle flat lay, or a premium packaging scene while keeping the label, shape, and color exactly as shipped. Generate a whole catalog of environments in minutes instead of booking a studio.

Campaign Creative from a Brief

Campaign Creative from a Brief

Feed GPT Image 2.5 a campaign brief with audience, key message, and tone, and generate the hero banner, social tiles, and product cards as a matching set. Multi-turn editing lets you adjust the headline or swap the hero image without losing the established visual direction.

Presentations and Infographics

Presentations and Infographics

Charts, growth callouts, slide layouts, and diagram-style visuals come out structured and readable. GPT Image 2.5 understands real-world information well enough to keep labels sensible and numbers legible, which makes it practical for decks, reports, and explainer graphics.

Lifestyle and Editorial Photography

Lifestyle and Editorial Photography

Candid street scenes, café portraits, interiors, and travel imagery with believable light and natural color. GPT Image 2.5 avoids the over-saturated, over-smoothed look, so lifestyle content for blogs, listings, and social feeds passes as photography.

How to Use

  1. 1

    Write a concrete prompt

    Describe the subject, setting, light, and any text you need. GPT Image 2.5 handles long, multi-clause prompts up to 20,000 characters, so include the details that matter rather than hoping the model guesses.

  2. 2

    Add reference images (optional)

    Drop in 1 to 16 reference images to preserve a face, product, or layout. GPT Image 2.5 switches to image-to-image automatically and keeps the recognizable features intact while applying your instructions.

  3. 3

    Choose tier, ratio and resolution

    Start with GPT Image 2.5 Flare for fast drafts, then switch to GPT Image 2.5 Sunburst for the final pass when precision matters. Pick one of 13 aspect ratios and 1K, 2K, or 4K output. Four narrow ratios (27:16, 16:27, 9:8, 8:9) are 1K only.

  4. 4

    Generate, then edit in place

    Review the result, then ask for targeted changes: recolor, replace, add text, remove an object. GPT Image 2.5 applies each edit without disturbing the rest of the image, so you can iterate to a finished asset in a few turns.

Try GPT Image 2.5 now

OpenAI's newest image model, live today. Flare for speed, Sunburst for polish, from 3 credits per image. Open Workspace and start generating.

Try in Workspace

Frequently Asked Questions

What is GPT Image 2.5?
GPT Image 2.5 is OpenAI's latest image generation and editing model, released on September 8, 2026 alongside ChatGPT Images 2.5. It improves on GPT Image 2 with sharper detail, more natural lighting and texture, stronger preservation of subjects from reference photos, more precise local edits, and better consistency across multi-turn editing. On this site you can use GPT Image 2.5 directly in the browser with no API key.
What is the difference between GPT Image 2.5 Flare and GPT Image 2.5 Sunburst?
They are two tiers of the same model with identical parameters. GPT Image 2.5 Flare is the default: it delivers the full quality upgrade at up to 50% lower latency than GPT Image 2, ideal for social content, thumbnails, prototypes, and high-volume work. GPT Image 2.5 Sunburst takes longer per image and is tuned for precision on detailed edits, polished product imagery, and campaign assets. A common workflow is to explore on Flare and finish on Sunburst. Both cost the same number of credits here.
How much does GPT Image 2.5 cost on this site?
Pricing is per image by resolution and is the same for Flare and Sunburst: 3 credits at 1K, 4 credits at 2K, and 7 credits at 4K. Text-to-image and image-to-image are priced identically, and the number of reference images does not change the cost.
Does GPT Image 2.5 support image-to-image and multiple references?
Yes. Upload from 1 to 16 reference images in JPEG, PNG, or WebP. GPT Image 2.5 uses them for identity preservation, style transfer, product placement, background replacement, and multi-image composition. Reference-based editing is where GPT Image 2.5 shows its largest lead over the previous generation.
What resolutions and aspect ratios are available?
GPT Image 2.5 outputs 1K, 2K, or 4K images in 13 aspect ratios: auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8, and 8:9. The four narrow ratios (27:16, 16:27, 9:8, 8:9) are available at 1K only; every other ratio supports 2K and 4K.
GPT Image 2.5 vs GPT Image 2: should I switch?
For most work, yes. GPT Image 2.5 keeps the text rendering and world knowledge that GPT Image 2 is known for, then adds better reference fidelity, cleaner local edits, more natural light, resolution tiers up to 4K, and faster generation on the Flare tier. GPT Image 2 remains available and uses a flat 4 credits per image, so it can still be the cheaper choice for quick 1K drafts.
GPT Image 2.5 vs Nano Banana Pro: which one should I pick?
Choose GPT Image 2.5 when you need a subject to stay recognizable across edits, when the brief includes text or layout, or when you are iterating on the same image over several turns; early Arena image-edit scores place both GPT Image 2.5 tiers above Nano Banana Pro. Choose Nano Banana Pro for its own strengths in 4K physical detail and fine-grained physics. Both are one click apart in Workspace, so run the same prompt on each and compare.
What are the known limitations of GPT Image 2.5?
Dense small type can still misrender, and text sometimes shows faint background patches. Characters may drift between separate generations (as opposed to edits within one session), and exact pixel placement in layout-heavy designs is not guaranteed. Some non-Latin scripts and multi-face group shots are weaker. Treat GPT Image 2.5 as a strong first draft plus editing tool, and use targeted edits to fix what it misses.

Also Try

Nano Banana
Google

Nano Banana

Generate stunning AI images with Nano Banana. Fast, versatile, and free to start. Create beautiful artwork, photos, and designs with text prompts.

Nano Banana Pro
Google

Nano Banana Pro

Professional AI image generation with Nano Banana Pro. Higher resolution, better detail, and advanced features for professional creators.

Nano Banana 2
Google

Nano Banana 2

The latest Nano Banana 2 model. More affordable, supports up to 14 reference images, and delivers excellent quality at lower cost.

Z Image
Tongyi

Z Image

Fast and affordable AI image generation with Z Image. Text-only generation at just 4 credits per image. Perfect for quick iterations.

GPT Image 2
OpenAI

GPT Image 2

GPT Image 2 is OpenAI's next-gen image model for in-image text rendering, pixel-level editing, and world-knowledge-driven realism. Supports text-to-image and image-to-image with up to 16 reference images.

Share this page
GPT Image 2.5 AI Image Generator — Flare & Sunburst Online | Nano Banana - AI Image Generator Platform