Updated: November 8, 2026
Image Model — generation & editing
UPDATE — November 8, 2026: GPT Image 2.6 Vector makes SVG files from a prompt: logos, icons and simple illustrations, in Flare or Sunburst at Low, Medium or High quality. 2.6 also covers text-to-image and editing (Low to High quality, Reasoning control, 1K to 4K, up to 16 references) and a GIF option for short loops.
This page covers the full GPT-Image-2 family: the original GPT-Image-2 release, GPT-Image-2.5 and GPT-Image-2.6.
GPT-Image-2 is OpenAI’s flagship image model, strongest on photorealism and on text that stays legible — signs, UI, packaging, infographics.
Since 2.5 it ships in two variants: Flare for everyday speed, Sunburst for precision work, both covering generation and editing.
Pick it over Nano Banana or Flux when a job depends on rendered text or on repeated edits to the same subject.
Variants at a Glance
Type |
Modalities |
Max Res. |
Talking Points |
|
|---|---|---|---|---|
| Variant | ||||
| GPT-Image-2 | Image | Text-to-Image, Editing | 4K | The original release, superseded by 2.5 and 2.6 for edit-heavy work |
| 2.5 Flare / Sunburst | Image | Text-to-Image, Editing | 4K | Proven pair: Flare for volume, Sunburst for precise, multi-turn edits |
| 2.6 Flare | Image | Text-to-Image, Editing | 4K | Default for most work: social, product, prototyping, high volume |
| 2.6 Sunburst | Image | Text-to-Image, Editing | 4K | Slower, tightest edit control: campaign creative, polished product imagery |
| 2.6 GIF | Animated GIF | Text-to-Image | 400px | Short looping animations from a prompt: stickers, reactions, small UI loops |
| 2.6 Vector | Vector (SVG) | Text-to-Image | Scalable | Logos, icons and simple illustrations as editable vector files |
Key Features
Image Editing
- Scoped Edits: Changes only what’s asked; subject, composition and background stay untouched
- Subject Consistency: Distinctive features carry through across new settings, styles and compositions
- Long Sessions: Quality holds over many rounds of revision instead of degrading
- Multi-Reference Input: Up to 16 reference images in one edit, each taggable in the prompt
Text-to-Image
- Text Rendering: Near-perfect text on signs, UI and infographics, the model’s standout strength
- Instruction Following: Handles detailed visual instructions and real-world information accurately
- Reasoning Control: On 2.6, choose how much the model plans before rendering; Medium by default, High for the most careful layouts
- Transparent Backgrounds: On 2.5, the background can be set to transparent or opaque for compositing work
- Extra Detail Tiers: On 2.5, XHigh and Max spend more effort on the image for extra detail
Animated GIF and Vector
- Prompt to GIF: One prompt becomes a short loop of 2 to 12 frames, timed from 20 ms to 2 seconds per frame
- Prompt to SVG: One prompt becomes a scalable vector file with clean flat shapes, ready to edit in design tools
Technical Capabilities
Details |
|
|---|---|
| Quality / Resolution | 2.6: Low, Medium, High, plus Reasoning (Low, Medium, High); 2.5: adds XHigh, Max; 1K, 2K, 4K (3840px long edge) |
| Aspect Ratios | 16:9, 9:16, 1:1, 2:3, 3:2, 3:4, 4:3 |
| Input Modalities | Text-to-Image, Image Editing |
| Reference Images | 1 to 16 per edit |
| Images per Run | 1 to 10 |
| Output Formats | PNG; SVG on 2.6 Vector; animated GIF on 2.6 GIF (2 to 12 frames, up to 400px); transparent background on 2.5 only |
| Variants | Flare (fast, default), Sunburst (precision, slower) |
Credit Cost
Credits are charged per image, or per frame for GIFs. Flare and Sunburst cost the same. On 2.6, editing costs more than text-to-image; on 2.5 the two cost the same.
GPT Image 2.6
| Mode · Quality | 1K |
2K |
4K |
|---|---|---|---|
| Text-to-Image · Low | 50 | 50 | 90 |
| Text-to-Image · Medium | 90 | 90 | 160 |
| Text-to-Image · High | 100 | 110 | 190 |
| Editing · Low | 70 | 70 | 110 |
| Editing · Medium | 100 | 110 | 180 |
| Editing · High | 120 | 120 | 200 |
GPT Image 2.6 Vector
| Quality | Credits per image |
|---|---|
| Low | 50 |
| Medium | 90 |
| High | 100 |
GPT Image 2.6 GIF
| Mode · Quality | Credits per frame |
|---|---|
| Text-to-Image · Low | 50 |
| Text-to-Image · Medium | 90 |
| Text-to-Image · High | 100 |
| Editing · Low | 70 |
| Editing · Medium | 100 |
| Editing · High | 120 |
GPT Image 2.5
| Mode · Quality | 1K |
2K |
4K |
|---|---|---|---|
| Text-to-Image and Editing · Low | 40 | 60 | 75 |
| Text-to-Image and Editing · Medium | 80 | 90 | 100 |
| Text-to-Image and Editing · High | 100 | 140 | 170 |
| Text-to-Image and Editing · XHigh | 150 | 200 | 250 |
| Text-to-Image and Editing · Max | 350 | 400 | 460 |
GPT-Image-2 (original)
| Quality | Credits per image |
|---|---|
| Low | 40 |
| Medium | 150 |
| High | 600 |
The original GPT-Image-2 at Low and Medium quality is included in Unlimited on the Creator, AI Professional and AI Pro Plus plans; High quality uses credits. GPT Image 2.5 and 2.6, including 2.6 GIF and 2.6 Vector, aren’t included in any Unlimited plan, so every generation uses credits. Credit costs can change; check the model in Artlist AI Tools for the current price.
Limitations
- Sunburst buys precision with generation time, so it runs slower than Flare
- 2.6 tops out at High; XHigh and Max are only on 2.5, and they add time on top of render time, so use them for finals, not for iterating
- Seven preset aspect ratios only; panoramas and thin banners have to be composed wider and cropped
- No mask or brush: edits are steered by the prompt alone, so name the region you want changed
- GIFs are prompt-only and top out at 400px, so they suit stickers and loops, not hero animation
- SVG is prompt-only and suits flat graphics; detailed scenes come out simplified and can take several minutes
- 2.6 always renders an opaque background, SVG included; use 2.5 for cutouts
- No seed control, so a prompt won’t reproduce the same image; lock a result by editing it forward
Prompting Tips
- Conversational full sentences beat keyword lists; the model reads a prompt the way a person would
- Put rendered copy in quotes and say where it sits: "the sign reads 'OPEN LATE' in bold condensed sans"
- For edits, name what should stay as well as what changes: "keep the background and lighting, replace only the label"
- With several references, tag each image in the prompt and say what it contributes: subject from one, setting from another
- For a GIF, describe the motion across the frames: "the cup fills from empty to full, then the steam rises"
- For SVG, ask for flat shapes and a small palette: "flat two-color logo of a coffee cup with steam"
- Iterate at Medium on Flare, then rerun the winner on Sunburst at High quality and High reasoning for the shipping frame
- For a cutout headed into a composite, generate on 2.5 and ask for a transparent background up front