Skip to main content
This page applies to the two GPT Image 2.6 official-relay models: gpt-image-2.6-sunburst (precision work) and gpt-image-2.6-flare (fast iteration). GPT Image 2.5 / 2 have no gif or svg output, so do not apply these rules to them directly.

​
Short answer

There is exactly one hard conflict: background: "transparent" cannot be combined with output_format: "jpeg"; it returns 400. Every other combination works, but the transparency semantics differ: Measured on GPT Image 2.6 official relay on 2026-11-12 (quality: "low", n: 1; GIF at 512×512 with 2 frames; SVG with svg.model: "gpt-4.1-mini"). The full matrix was measured on gpt-image-2.6-sunburst; jpeg + transparent, gif + transparent, and svg + opaque were re-checked on gpt-image-2.6-flare with identical results.

​
JPEG is the only 400

JPEG has no alpha channel, so background: "transparent" fails with:
This applies to both /v1/images/generations and /v1/images/edits. Switch to png / webp (or gif for animation); opaque and auto work fine with JPEG.
On the edit endpoint this 400 took about 62 seconds to return in our test (about 19 seconds on text-to-image), so do not retry just because the request seems slow.

​
GIF supports transparency, but only 1-bit alpha

output_format: "gif" + background: "transparent" works on 2.6, on both text-to-image and image editing. Transparent pixels covered 46%–60% of the frame in our tests, versus 0% with opaque. GIF transparency is palette-based: a pixel is either fully transparent or fully opaque, with no soft alpha edges. Use png / webp when edge quality matters. This only applies to 2.6, which supports GIF output; 2.5 and earlier do not support GIF output.

​
SVG has no alpha; opaque adds a background

SVG is a vector format with no alpha channel, and transparency is the default:
  • transparent: no background rectangle is written.
  • opaque: a full-canvas white rectangle is added, such as <rect width="1024" height="1024" fill="#fff" />.
  • auto: in both of our tests, auto wrote no background, so do not treat it as opaque; pass opaque explicitly when you need a white background.
output_format: "svg" is supported only on the text-to-image endpoint, requires n: 1, and requires svg.model. The author model decides the final background markup, so re-check the actual file after changing svg.model.

​
auto is not fixed to opaque

background defaults to auto when omitted. auto follows the prompt rather than always producing a solid background:
  • With a prompt that asks for a plain white background, png / jpeg / gif reported opaque and had no alpha.
  • With a prompt that asks for an isolated subject on no background, png / webp / gif / svg reported transparent; the first three actually had transparent pixels.
  • With jpeg, auto is forced to opaque and does not error.
For deterministic output, pass transparent or opaque explicitly instead of relying on auto.
  • output_compression only applies to jpeg / webp.
  • frame_count / frame_duration_ms are only for gif; do not send them to PNG, JPEG, WebP, or SVG.

GPT Image 2.6 capabilities

SVG, GIF, complexity, and markup-guided editing with real results.

How to generate transparent images

Parameter shape, prompt requirements, and common errors for transparent backgrounds.

Text-to-image API reference

Every field for /v1/images/generations.

Image editing API reference

/v1/images/edits and multi-image editing.