What is ducttape?

"ducttape" (also written "duct-tape" or "duct tape") was the anonymous handle GPT Image 2 tested under on LMArena from late 2025 into early 2026. It has now been confirmed as OpenAI's next-generation image model — the successor to GPT-Image-1.

On the Arena interface it appeared as numbered variants — duct-tape 1, duct-tape 2, duct-tape 3 — representing different model versions or settings. Community consensus is that duct-tape 2 produced the most balanced output across prompts.

Three properties showed up in almost every side-by-side post. First: text rendering. Posters, menus, signage, packaging mockups, album covers, and brand moodboards came out with legible, properly kerned type — still a rare capability in image models as of early 2026.

Second: layout fidelity. When a prompt specified a composition ("three panels, the third one is vertical"), GPT Image 2 actually obeyed it. When a user supplied a reference image, the output preserved angle, crop, and palette more faithfully than Nano Banana Pro or Seedream 4 in the same prompt.

Third: multilingual text. Korean, Japanese, and Chinese characters rendered correctly rather than as decorative nonsense. This was the single strongest signal to non-English communities that something unusual was in testing.

The pattern is now familiar. In mid-2025 the codename nanobanana was tested on LMArena before Google confirmed it as Gemini 2.5 Flash Image. Sites that tracked the codename early captured disproportionate organic traffic when the reveal landed.

ducttape followed the same arc — months of anonymous Arena testing, growing community attention, and now confirmation as GPT Image 2. The pre-reveal window is closed; the launch window is open.

ducttapeai.app tracked the codename from day one. Now that ducttape is confirmed as GPT Image 2, this site is transitioning into a generation tool. When OpenAI opens the GPT Image 2 API, you will be able to generate images here. Subscribers hear the moment it goes live.