OpenAI brings transparent background support to gpt-image-2 in preview

OpenAI has added native transparent-background output to gpt-image-2, its flagship image generation model, according to an entry posted to the OpenAI API changelog on August 20, 2026. The capability is available in preview across both the Images API and the Responses API's image generation tool.

What's new

The changelog states plainly: "Transparent backgrounds are now available in preview for gpt-image-2 and gpt-image-2-2026-04-21 in the Images API and the Responses API image generation tool." Developers can now request an alpha-channel PNG or WebP output directly from the model rather than generating an image on a solid background and removing it in a separate post-processing step.

The feature applies to two specific model identifiers — the general gpt-image-2 endpoint and the dated snapshot gpt-image-2-2026-04-21 — meaning integrations pinned to either version should be able to pick up transparency support without a model swap. Because the capability ships through both the standalone Images API and the newer Responses API's built-in image tool, it reaches both classic image-generation integrations and agentic workflows that call image generation as a tool mid-conversation.

Context

Transparent-background generation has historically been a weak point for diffusion- and autoregressive-style image models: most default to filling in a background rather than leaving it empty, forcing developers to chroma-key or matte the subject out afterward — a lossy, fiddly step for logos, product cutouts, stickers, and UI assets. gpt-image-2 launched as OpenAI's default image model, and third-party workflows have generally leaned on rival tools when a clean alpha channel was the hard requirement, since earlier gpt-image versions could not reliably produce one natively.

The "preview" label signals this is an early-access capability rather than a finished, generally-available feature — OpenAI has historically used preview releases to gather developer feedback before locking in output quality guarantees or pricing terms for a capability.

Why it matters

Native transparency removes an entire post-processing step from a common production pipeline: brand asset generation, e-commerce product photography, sticker and asset packs, and UI icon work all depend on a clean cutout. Shipping it directly into gpt-image-2 rather than as a separate tool also keeps that workflow inside OpenAI's own API surface, competing more directly with tools purpose-built for background removal and transparent asset generation. For teams already standardized on gpt-image-2 for other image work, this closes a capability gap that previously sent transparent-asset requests to a different model or vendor. As with any preview feature, output consistency (edge cleanliness, alpha-channel fidelity) is worth validating against a given use case before relying on it in a production pipeline.