OpenAI adds transparent background generation to GPT-Image-2 API preview

OpenAI said on August 21 that GPT-Image-2 now supports transparent background generation in preview through its API, giving developers a way to return images with an alpha channel by setting the background parameter to "transparent." The update restores a feature that had been missing from GPT-Image-2 since its April 21 release and early-May developer rollout, even though the older gpt-image-1 had supported transparent output since April 2025. Before this preview, developers had to rely on workarounds. Some downgraded to older models such as gpt-image-1.5 or sora_image. Others added background-removal tools like RemBG, BiRefNet, or 851-labs' background-remover, adding 1 to 3 seconds per image. More complex methods included generating green-screen backgrounds or creating the same image twice on different backgrounds and subtracting the results, which doubled generation costs. The report also outlined GPT-Image-2 pricing at roughly $0.03 for 1K, $0.05 for 2K, and $0.08 for 4K images. For a 1024×1024 image, output can range from 272 tokens at low quality to 4,160 tokens at high quality, with output priced at $30 per million tokens. The feature remains in preview and comes with limits. Early testers said alpha values landed at 253 to 254 instead of a fully opaque 255, edge halos could appear around subjects, and the edit endpoint might not preserve transparency from source images. Transparent output is currently limited to PNG and WebP, while OpenAI has not disclosed any preview-specific pricing differences or usage caps.