GPT Image 2.6 in WinkImg — the model, its two variants, and what it changes
WinkImg generates images with GPT Image 2.6 — the model OpenAI shipped on November 4, 2026. This page explains what the model is, how its two variants differ, and how it behaves when you drive it through a fixed style template instead of a free-form prompt.
GPT Image 2.6 replaced the previous ChatGPT Images model in ChatGPT, Codex and the API on November 4, 2026. The release is less about a new look than about three practical changes: generation got roughly 50% faster, edits hold character and product details more consistently across rounds, and the model can now output animated GIFs and accept photo markup.
Developers get it as two model names rather than one. GPT-Image-2.6 Flare is tuned for speed — a frame can come back in about six seconds. GPT-Image-2.6 Sunburst is tuned for consistency — the same face, product and palette stay put while you keep editing. They share a prompt format and output shape, so switching between them does not change how you write.
That split matters for a template-based tool. A style template is a fixed lighting/material/composition recipe, so the interesting question is not "can the model draw" but "can it hold the recipe across many images in one series". Flare gets you the frame fast enough to iterate; Sunburst is what keeps a series looking like one series.
Flare vs Sunburst, side by side
| Dimension | GPT-Image-2.6 Flare | GPT-Image-2.6 Sunburst |
|---|---|---|
| Time to first frame | About six seconds for a typical prompt. Fast enough to try two or three phrasings in a row. | Slower per frame, because it spends more work holding reference details. |
| What it optimises for | Throughput. Everyday stills, quick variations, GIF frames. | Consistency. Same character, same product, same brand details across edits. |
| Best with a style template | Exploration: you keep the template fixed and change only the subject line. | Series work: one character or product rendered across many styles or angles. |
| Where it shows weakness | Long edit chains can drift — the fifth edit looks less like the first. | Cost and latency. Not the model to use for rapid trial and error. |
How WinkImg actually uses it
Flare is the default path
Every generation on WinkImg goes through GPT-Image-2.6 Flare by default, at low, medium or high quality. The template supplies the look; Flare supplies the speed, which is why a one-line prompt comes back in seconds rather than a minute.
25 templates, one model
The 25 style templates are not 25 different models. They are 25 tuned recipes — figurine, anime, pixel art, cyberpunk, ink wash and so on — pushed through the same model. That is deliberate: one model means the material and lighting behaviour stay predictable when you move from style to style.
GIF comes from the same model
Because 2.6 can output GIFs, a style can become a short looping clip instead of a still. WinkImg renders 2 to 16 frames and packs them into one sheet before encoding, so the loop stays consistent frame to frame instead of flickering.
What it does not do
We do not offer a model picker today. You cannot select Sunburst, and we do not claim to. If a task needs Sunburst-grade consistency, the honest answer is that this page is documentation, not a switch.
Boundaries, stated up front
- —WinkImg is not affiliated with OpenAI and does not carry an OpenAI logo. "GPT Image 2.6" is used here to describe the underlying model, the same way a laptop page names its CPU.
- —Free generations reset daily and cover the standard quality tier. Higher quality and GIF output draw on the paid credit balance.
- —Model naming can change upstream. When OpenAI ships 2.7, this page gets a sibling at /models/gpt-image-2-7 and stays online — we do not redirect old model pages, because pages with history are worth keeping.
Questions people actually ask
+Is GPT Image 2.6 the same as ChatGPT Images 2.6?
Yes. OpenAI markets the release as Images 2.6 inside ChatGPT and ships it to developers as GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Same release, two names depending on where you use it.
+Which variant does WinkImg use?
Flare by default, across all three quality tiers. It is the variant tuned for speed, which is what a one-line-prompt tool needs. Sunburst is not exposed as a user-facing option.
+Can I use GPT Image 2.6 for free here?
You can generate free images every day after signing in with Google — standard quality tier, no watermark, no credit card. Higher quality tiers and GIF output use purchased credits.
+Does 2.6 fix the consistency problem across a series?
Partly. Within a fixed template, material and lighting stay stable because the recipe is fixed. Subject drift across many rounds of edits is reduced compared with earlier image models, but it is not eliminated; changing one detail at a time still beats rewriting the whole line.
+What is the GIF limit?
2 to 16 frames per GIF. All frames are generated on one sheet and then encoded, which is why the loop does not flicker between frames.
+Will there be a page for GPT Image 2.7?
When we actually run it. We add a model page only for models in production here, not for every model that ships.
A few styles rendered through this model
Try GPT Image 2.6 through a style template