gpt-image-2.6 GIF and SVG Output: Parameters and Limits

OpenAI released GPT-Image-2.6 (Sunburst and Flare) on November 4, 2026. For the first time, the Image API can return animated GIFs and SVG vector files directly. This guide summarizes the parameters and limits from OpenAI's image generation guide so you can plan an integration before writing code.

Short answer: set output_format to gif or svg, but both are generation-only

Send POST /v1/images/generations with model set to gpt-image-2.6-sunburst or gpt-image-2.6-flare and output_format set to "gif" or "svg". GIFs accept 2 to 16 frames at 20 to 2000 ms per frame. SVG returns one image per request. Neither format is available from the edit endpoint.

Animated GIF parameters

Describe the subject, the motion and how the loop should return to the start. Three fields control the animation.

{"model": "gpt-image-2.6-flare", "prompt": "A steaming coffee cup; the steam rises slowly and returns to the first frame to loop", "output_format": "gif", "frame_count": 9, "frame_duration_ms": 200, "size": "auto"}

Decode the returned b64_json and save it with a .gif extension.

Loop length = frames x frame duration

OpenAI's own example uses Sunburst with frame_count=9 and frame_duration_ms=200, and the character returns to its starting pose after 1.8 seconds. Start with the defaults, then add frames if the motion looks choppy.

Five GIF limitations

• Generation only: /v1/images/edits does not return GIFs, so you cannot animate an existing photo.
• Pixel budget: frame count, frame size and n must fit the model's size limits and the total request pixel budget. Reduce one of them if the request fails.
• No streaming or partial images.
• No custom output_compression.
• No prompt_rewriter.

SVG output

With output_format "svg", you get editable paths, shapes and colors, which suits icons, logos and illustrations.

How SVG works and how it is billed

The API first creates a reference image and then authors the SVG from it; quality applies to that reference image. You pay the image model's rate for the reference image plus the authoring model's rate for the tokens used to write the SVG (Cost and latency). Standard token rates for 2.6 match 2.5 ($30 per 1M image output tokens), but 2.6 does not support the Batch API, while 2.5 does at $15.

SVG limitations

• n must be 1.
• No streaming, partial images, output_compression or prompt_rewriter.
• No edits: the edit endpoint cannot output SVG, and SVG is not a valid edit input.
• Static content only: no scripts, external resources or embedded raster images.

Also new: markup-guided edits

You can now send a transparent PNG with circles or sketches as markup to /v1/images/edits to show where a change should happen. This also works on supported GPT Image 2.5 models. The first input image and the annotation layer can each be up to 16,777,216 pixels, the annotation file up to 50 MB, and markup cannot be combined with mask or prompt_rewriter. There is no extra fee beyond normal edit pricing.

Which format for which job

If you want to build image generation into an internal workflow or product, we help teams design and build custom AI integrations.

Summary

With gpt-image-2.6, setting output_format to gif or svg is all it takes to get an animated GIF or a vector SVG from the API. GIFs take 2 to 16 frames at 20 to 2000 ms each, and loop length is the product of the two. Both formats are generation-only. Start with the default 9 frames at 200 ms and tune from there.