Skip to content

GPT-Image โ€‹

Updated

Endpoint: POST https://vod.intl.tencentcloudapi.comAction: CreateAigcImageTask

GPT-Image is a family of image generation models. It supports text-to-image and multi-reference image generation, plus mask-based local editing, transparent backgrounds, multi-image output, and custom sizes; the 2.6 series can also output animated GIFs and SVG vector images directly. Multilingual text rendering is strong, which makes it a good fit for posters and type-heavy design work.

Three series are available today: image2, image2.5, and image2.6. All three share the same ModelName=OG and the same request fields โ€” for basic usage only the ModelVersion value differs.

Choose a version โ€‹

The API reference is split per series. Open a card to see that series' ModelVersion enumeration, quality tiers, generation times, and special settings.

Version comparison โ€‹

Item2.02.52.6
ModelNameOGOGOG
ModelVersionimage2_low
image2_medium
image2_high
10 values
five tiers each for sunburst / flare
10 values
five tiers each for sunburst / flare
Variants12 (sunburst / flare)2 (sunburst / flare)
Quality tierslow / medium / highplus xhigh / maxplus xhigh / max
ModelVersion values31010
Multi-reference (up to 16)โœ…โœ…โœ…
mask inpaintingโœ…โœ…โœ…
Transparent backgroundโœ…โœ…โœ…
Custom size and autoโœ…โœ…โœ…
Multi-image output (1โ€“8)โœ…โœ…โœ…
Output format OutputFormatjpeg / pngjpeg / pngjpeg / png
Animated GIF (2โ€“16 frames)โŒโŒโœ…
SVG vector imageโŒโŒโœ… (extra SVG conversion fee)

All three series share one set of request fields

FileInfos, OutputConfig, and ExtInfo are written identically across the three series. Switching series means changing ModelVersion only โ€” the request structure stays the same. The one exception is GIF / SVG output: it is enabled with output_format in ExtInfo and only 2.6 supports it.

Choosing a version โ€‹

  • Prefer 2.6 for new integrations: it is the latest generation and its basic parameters match 2.5, so switching costs nothing.
  • Animated GIF or SVG output requires 2.6: set output_format=gif / svg in ExtInfo; see GIF and SVG output. Note that SVG carries an extra conversion fee.
  • Pick flare when latency matters: at the same tier, flare is clearly faster than sunburst (measured on 2.5: 61 s vs 169 s at the max tier).
  • Pick sunburst for editing precision, complex compositions, and small text, and raise the tier to xhigh or max.
  • Use low / medium for quick drafts and prompt iteration; the 2.0 series is sufficient here and has the simplest parameters.
  • If three tiers are enough, the 2.0 series is still available โ€” it is the default series when ModelVersion is omitted (image2_medium).

Shared specifications โ€‹

The following applies to all three series. Details and examples live in the per-version documents.

ItemValue
DefaultsModelVersion=image2_medium / AspectRatio=1:1
Resolution Resolution1K / 2K (2560x1440) / 4K (3840x2160); anything above 2560px is experimental
Aspect ratio AspectRatio1:1 / 3:2 / 2:3 / 3:4 / 4:3 / 16:9 / 9:16 / 21:9 / 9:21
Reference imagesUp to 16
Prompt lengthUp to 32,000 characters
Text renderingChinese, English, Japanese, Korean, Arabic, and more
Multi-image output1โ€“8 for OG (1โ€“9 for Kling)
Reference image inputFile (FileId) / Url / Base64; Base64 total โ‰ค 7 MB, formats jpeg / jpg / png / webp, no data: prefix
Custom sizeWidth and height both multiples of 16; longest edge โ‰ค 3840px; total pixels 655,360 โ€“ 8,294,400
Guardrail bypassNot supported
DocumentDescription
GPT-Image 2.6Latest series: five tiers each for sunburst / flare, GIF / SVG output
GPT-Image 2.5Two variants, xhigh / max tiers, measured generation times
GPT-Image 2.0Base series, quality tiers up to high
VOD quick startAuthentication, task submission, and result polling for the VOD API

Built with VitePress ยท Deployed on Tencent EdgeOne Pages