Skip to main content
2026/8/21 16:57 (UTC+8) · New Model · OpenAI 🚀 gpt-image-2 now does transparent backgrounds — one parameter gets you a real alpha-channel PNG, no manual cutouts OpenAI opened up the transparent value of the background parameter for GPT-Image-2 today (marked preview by OpenAI), and APIYI has verified it works. Two fields in the request is all it takes: "background": "transparent" plus "output_format": "png", and what comes back is a PNG with a real alpha channel. webp also carries alpha and produces smaller files. All three paths work: text-to-image /v1/images/generations, image editing /v1/images/edits, and the image_generation tool on /v1/responses. Mask inpainting (mask) and a transparent background work together without conflict. Transparency costs nothing extra — at the same quality tier and size, transparent and opaque consume exactly the same image tokens. Two boundaries worth knowing:
  • jpeg has no alpha channel and is mutually exclusive with transparency; passing output_format: jpeg returns 400, so use png or webp
  • On the edit endpoint, transparency is a re-draw, not a precise trace of the original outline — subject detail will shift. For pixel-exact extraction, run rembg / PIL / sharp yourself
The reverse models gpt-image-2-all and gpt-image-2-vip have no background parameter — you can only ask for transparency in the prompt, and it is occasionally unreliable. For dependable transparent backgrounds, use official-relay gpt-image-2. Per-model support, minimal examples, and common errors: How do I generate images with a transparent background.
← Back to Live Updates · 📚 Monthly Archive