GPT Image 2.6 Guide: Animated GIFs, SVG, Markup Edits and Transparent Cutouts for Marketing

Marketing teams have asked image models for the same three things for years: assets that move, assets that scale, and edits that touch only what you point at. GPT Image 2.6 is OpenAI's first image release that goes after all three in one model family, and it is listed from November 4, 2026.

This guide separates what OpenAI, fal and independent reviewers document from what you can do in MITPO Creative Studio today, and shows how to brief each new format so the first render is usable.

Key Takeaways (AEO Direct Summary)

• Two models: GPT Image 2.6 ships as Flare (fast, everyday generation) and Sunburst (detail and edit consistency). In MITPO they appear as "GPT Image 2.6" and "GPT Image 2.6 Precise".
• Animated GIFs: a prompt can return a short looping animation. fal's schema allows 2 to 16 frames at 20 to 2,000 ms per frame, with 9 frames at 200 ms (a 1.8 second loop) as the default.
• SVG output: one static vector file per request, drawn from text alone. It cannot trace an existing logo.
• Markup edits: a transparent layer of circles, arrows and notes tells the model exactly where an edit goes.
• In MITPO today: still images at 6 credits per image (11 at 4K) on Starter and above. GIF, SVG and markup are API features we are testing before we switch them on in Creative Studio.

What GPT Image 2.6 adds

• New output formats: GIF and SVG join PNG, JPEG and WebP. SVG is text-to-image only.
• Markup-guided edits: alongside reference images and a mask, you can draw a transparent markup layer over the first image.
• A planning control: low, medium or high planning before rendering (OpenAI calls it complexity). High is the default.
• More quality levels: low, medium, high, xhigh and max in fal's schema.
• Speed: Maverick Frame's summary of the release says Flare can answer in as little as about 6 seconds at low quality.

For everything 2.5 already did, such as transparent backgrounds, long prompts and reference edits, see the GPT Image 2.5 guide.

Both tiers are listed at the same token rates on OpenAI's model pages ($5 per million text input tokens, $8 image input, $30 image output), so the choice is about the job, not the price. Output tokens are billed, which means size and quality move the cost. One developer who tested the release reported more output tokens per image than 2.5 at the same setting, so measure a sample before you plan a large batch.

Flare or Sunburst?

Use Flare to find the idea, then move to Sunburst to refine it.

Animated GIFs: one simple action

A GIF is a loop, so the brief is different from a still. The pattern that works, and the one Higgsfield recommends in its own guide, is one simple action with a fixed camera: a wave, a blink, a pulsing light.

Three constraints shape every GIF brief:

• All frames share one pixel budget. A longer loop has to be smaller, so you cannot max out both frame count and size.
• GIFs come from the generation request, not the edit request. You cannot animate a photo you already have.
• Frame limits disagree. OpenAI's guide mentions up to 64 frames while its API reference and fal's schema say 16. Plan for 16 or fewer.

A template to adapt:

Flat-colour illustration with a clean uniform outline, fixed camera.
[Object] sits centred on a plain [colour] background.
Looping animation of one simple action: [what moves and how].
The first and last frame show the identical resting pose so the loop is seamless.
The design, colours, outline weight and framing are identical in every frame;
only [the moving part] changes. No text.

Good uses: a pulsing notification dot for an email header, a ringing phone for a support page, a steaming cup for a coffee brand's story. Check the first and last frame side by side before you ship, because a loop that jumps at the seam reads as a glitch.

SVG: scalable icons and simple mascots

SVG output returns a static vector file you can scale to a billboard or shrink to a favicon. The documented limits matter for planning:

• It is text-to-image only: there is no reference image, so it cannot vectorize your existing logo or artwork.
• One file per request, with no scripts, no outside resources and no embedded raster images.
• It is billed in two parts: the image model's rate for a reference image it renders first, plus the rate of the model that writes the SVG.

Use it for icons, simple emblems and mascots you are creating from scratch. If a brand mark already exists, trace or redraw it instead.

Markup edits: point instead of describe

Describing an edit in words is the slow part of every revision round: "the second card, no, the one on the right, make the corner rounder". Markup replaces that. You draw circles, arrows and short notes on a transparent layer the same size as your image, and the model reads them as instructions for the first image. It cannot be combined with a mask.

A practical workflow for packaging, layouts and product shots:

1. Start from a rough sketch or a simple 3D blockout, so composition is already locked.
2. Draw a circle around each part and write the material or change next to it.
3. Say in the prompt that the layer is instructions and none of the marks should appear in the result.
4. Run the edit at a low planning level. OpenAI's own guidance is low for most edits, because higher settings can introduce unwanted changes.

Finish the first image as a photoreal product photograph.
The second image is a transparent layer of circles, arrows and handwritten notes:
follow each note for the part it points at, and keep none of the markup in the final picture.
Keep the layout, camera angle and spacing exactly as they are.

Transparent cutouts

GPT Image 2.6 keeps the background setting from 2.5: auto, transparent or opaque. Choose transparent when you need a product on a changing backdrop, a sticker, or a story overlay. The test of a good cutout is the hard edge: fine hair, fabric fuzz, glass and water droplets. Ask for "a clean alpha edge that keeps every fine detail" and check the result on a checkerboard, not on white.

What to know before you plan around it

• Batch is not available for 2.6, and there is no dated snapshot to pin yet. Keep large overnight jobs on 2.5.
• Planning defaults to high. Set it deliberately: low for edits, higher for scenes with many objects.
• Transparency, GIF, SVG and markup need an app that switches them on. Creative Studio renders still images today; we are testing the new formats before we expose them.
• Always read printed text letter by letter, whichever model made it.

Using GPT Image 2.6 in MITPO

Open Creative Studio, choose GPT Image 2.6 for first drafts or GPT Image 2.6 Precise for edits, and attach reference images from the prompt bar. Both are 6 credits per image at standard size and 11 at 4K, and both are available on Starter and above. See pricing for plan details, the Creative Studio docs for the controls, and the GPT Image 2.5 guide for how the previous generation compares.

FAQ

What is GPT Image 2.6?

It is OpenAI's image model family released on November 4, 2026, in two tiers: Flare for speed and Sunburst for detail and edit consistency. It adds GIF and SVG output, markup-guided edits and a planning control to what 2.5 offered.

Can GPT Image 2.6 make animated GIFs?

Yes, on the generation request. You set a frame count and a frame duration; all frames share one size budget, so longer loops need smaller frames. GIFs are not available on the edit request. MITPO does not expose GIF output yet.

Can GPT Image 2.6 turn my logo into an SVG?

No. SVG output draws a new vector from text and accepts no reference image, so it cannot convert existing artwork. Use it for new icons and simple emblems, and redraw approved brand marks.

What is the difference between Flare and Sunburst?

Flare is the faster tier for layouts and quick variations. Sunburst is built for detail and for keeping everything you did not mention unchanged across edits. In MITPO they are GPT Image 2.6 and GPT Image 2.6 Precise.

How much does GPT Image 2.6 cost in MITPO?

6 credits per image at standard size and 11 at 4K, on Starter and above. The model picker always shows the live cost before you generate.