OpenAIGPT Image 2.5
Flare and Sunburst
OpenAI's newest image model, built for sharper detail, more faithful reference photos, and edits that change only what you asked for. Two variants ship in the API: Flare for fast everyday work at up to 50% lower latency than GPT Image 2, and Sunburst for premium creative workflows that need tighter control across edits.
Both variants are live on faktry. Sign up and you get 100 free credits, no credit card required.
Made with GPT Image 2.5
Text rendering, layout control, and style fidelity in a single pass.
What is GPT Image 2.5?
GPT Image 2.5 is OpenAI's image generation and editing model, released on 8 September 2026 alongside ChatGPT Images 2.5. It produces more natural lighting and texture than GPT Image 2, preserves the subjects in reference photos more reliably, and follows editing instructions more consistently across multiple turns. Two variants are available through the API: GPT-Image-2.5 Flare, the default for most work, and GPT-Image-2.5 Sunburst, aimed at premium creative and editing workflows. Both accept prompts up to 32,000 characters, render transparent backgrounds, and take up to 16 reference images per edit.
Key facts
- Developer
- OpenAI
- Released
- 8 September 2026
- Variants
- GPT-Image-2.5 Flare, GPT-Image-2.5 Sunburst
- Modalities
- Text to image, image editing with references
- Quality tiers
- low, medium, high, xhigh, max
- Sizes
- 1024x1024, 1536x1024, 1024x1536 and custom dimensions
- Reference images
- Up to 16 per edit
- Prompt length
- Up to 32,000 characters
- Transparent background
- Supported, on PNG and WebP output
- Provenance
- C2PA metadata and invisible watermarking
Last updated: · Source: OpenAI
Flare or Sunburst?
Same price per image, same quality tiers. The difference is speed against precision.
GPT-Image-2.5 Flare
Higher quality than GPT Image 2 at up to 50% lower latency. Built for creator and social content, product experiences, visual search, rapid prototyping, and high-volume generation.
Best for: everyday generation, fast iteration, large batches
GPT-Image-2.5 Sunburst
Built for premium visual workflows that benefit from tighter control across edits, at the cost of longer generation times. Use it for production-ready campaign creative and polished product imagery.
Best for: final assets, detailed edits, brand-critical work
Both variants cost the same on faktry. Pick Flare unless an edit needs the extra control.
GPT Image 2.5 vs GPT Image 2
What actually changed between the two generations.
| Capability | GPT Image 2 | GPT Image 2.5 |
|---|---|---|
| Variants | Single model | Flare (fast) and Sunburst (precise) |
| Generation speed | Baseline | Flare runs up to 50% lower latency, measured at two to four times the speed in partner evaluations |
| Quality tiers | low, medium, high | low, medium, high, plus new xhigh and max |
| Reference fidelity | Reads every input at high fidelity | Same high-fidelity reading, with subjects, lighting and distinctive features carrying through more reliably |
| Precision editing | Edits can drift into untouched areas | Better at changing only what was asked while preserving composition and surroundings |
| Multi-turn edits | Quality can degrade across repeated edits | Earlier changes hold up and quality is maintained across turns |
| Transparent backgrounds | Not supported | Supported, with improved results on PNG and WebP |
| Cost on faktry | Flat 10 credits | 1 to 40 credits depending on the quality tier you pick |
What it costs
GPT Image 2.5 is billed by quality tier, so drafts stay cheap and only final renders cost real credits. Prices below are for a square image; portrait and landscape frames are around 20% cheaper.
| Quality | Credits | Use it for |
|---|---|---|
| low | 1 | Thumbnails, layout tests, quick drafts |
| medium | 2 | Concepting and internal review |
| high | 10 | The default. Finished work for most uses |
| xhigh | 20 | Detail-critical assets, large prints |
| max | 40 | Maximum detail, when nothing less will do |
Editing starts at 8 credits plus 2 per reference image, so an ordinary one-reference edit at high quality lands at 10 credits.
What GPT Image 2.5 does well
The improvements that show up in real work, not just benchmarks.
Faithful to your references
Subjects stay recognisable across new settings, styles and compositions. Lighting and textures read as natural, and distinctive features carry through instead of being smoothed away.
Edits only what you asked
Change a product, a background or a line of copy while the subject, composition and brand treatment around it stay put. This holds up on complex subjects and busy backgrounds.
Holds up across turns
Each edit builds on the last instead of degrading it. Earlier changes stay consistent, so a long refinement session ends somewhere useful.
Complex layouts and text
Better at reading complex visual instructions and turning them into coherent results, including images that carry real-world information and layouts with transparent backgrounds.
Style that does not drift
Requested visual direction, composition and individual details are retained as briefs get more specific, which matters for a series of on-brand assets.
Fast enough to iterate
Flare cuts latency by up to half against GPT Image 2, so you can generate, look, adjust and go again without losing the thread.
Both operations, one model
GPT Image 2.5 handles generation and editing under the same model ID.
Text to image
Image editing
What people build with it
Where the reference fidelity and edit control actually pay off.
Campaign creative
Production-ready advertising visuals that hold a defined art direction across a whole set of assets.
Product imagery
Swap a background, a colourway or a label while the product itself stays exactly as photographed.
Social and creator content
High-volume generation at speed, with Flare turning a rough idea into a usable post in seconds.
Presentations and UI concepts
Slide visuals and interface mockups that respect a given hierarchy and structure instead of drifting off brief.
Can I try GPT Image 2.5 for free?
Yes. Sign up and 100 credits land in your account immediately, no card required.
Free credits to start
Frequently asked questions
What is GPT Image 2.5?
GPT Image 2.5 is OpenAI's image generation and editing model, released on 8 September 2026. It brings sharper detail, more natural lighting and texture, better preservation of subjects from reference photos, and more reliable editing across multiple turns than GPT Image 2. Two variants are available to developers: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
Can I try GPT Image 2.5 for free?
Yes. Sign up at faktry.ai and you get 100 free credits instantly, with no credit card required. A low-quality GPT Image 2.5 image costs 1 credit and a standard high-quality one costs 10, so the free credits go a long way. After that there is no subscription, you pay only for what you generate.
What is the difference between Flare and Sunburst?
Flare is the default for most applications. It delivers higher quality than GPT Image 2 at up to 50% lower latency and suits creator content, product experiences, prototyping and high-volume work. Sunburst is built for premium visual workflows that need tighter control across edits, such as campaign creative or polished product imagery, and takes longer to generate. They cost the same per image on faktry, so choose Flare unless an edit needs the extra precision.
How is GPT Image 2.5 better than GPT Image 2?
GPT Image 2.5 generates faster, preserves reference subjects more reliably, edits only what you asked for while leaving the rest of the image intact, and holds quality across repeated edits instead of degrading. It also adds two new quality tiers above high, xhigh and max, and supports transparent backgrounds, which GPT Image 2 does not.
When was GPT Image 2.5 released?
OpenAI announced ChatGPT Images 2.5 and the two API models, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst, on 8 September 2026. Both variants were available on faktry the same week.
Which quality setting should I use?
Use high for finished work, it is the default and matches what most people expect from an OpenAI image. Drop to low or medium for drafts, thumbnails and layout tests, where a render costs 1 or 2 credits instead of 10. Reach for xhigh or max only when the extra detail is genuinely the point, since they cost 20 and 40 credits respectively.
How many reference images can I use?
Up to 16 per edit. GPT Image 2.5 reads every reference at high fidelity, which is what makes subjects carry through so reliably, but it also means each one adds to the cost. Editing starts at 8 credits plus 2 per reference, so pass the references you actually need rather than everything to hand.
Can I use GPT Image 2.5 images commercially?
Yes. Images you generate on faktry can be used commercially, including marketing, advertising, client work and products. Note that OpenAI embeds C2PA metadata and invisible watermarking in its output to identify images made with its tools.