ducttapeai.app
GPT Image 2 · OpenAI · Coming soon

ducttape is GPT Image 2. Generation is coming to this site.

The anonymous image model that dominated LMArena under the codename ducttape is OpenAI's GPT Image 2 — the successor to GPT-Image-1. Text renders like a designer typed it. Layouts hold. Multilingual output works. When the API goes live, you'll be able to generate here. Subscribe to be first.

One email when GPT Image 2 generation goes live here. No spam, no marketing — we unsubscribe you ourselves if we ever break that rule.

GPT Image 2 samples — 100% AI-generated

Every image below was generated by GPT Image 2 (ducttape) on LMArena. No post-processing, no cherry-picking.

Source: @arrakis_ai

What GPT Image 2 does best

Text renders like a designer typed it

Posters, menus, packaging, multi-line captions — GPT Image 2 keeps letters on the baseline, kerning intact, and misspellings rare. The single most praised capability on LMArena.

Reference images actually follow

Given a reference, GPT Image 2 preserves composition, pose, and palette more faithfully than Nano Banana Pro or Seedream 4 in head-to-head pairings.

Multilingual output is stable

Korean, Japanese, and Chinese characters render correctly — not romanised, not mangled. Rare for any image model as of early 2026.

How GPT Image 2 compares

Blind side-by-side results from LMArena against the leading image models. Based on community votes from thousands of anonymous pairings.

Google · Gemini 2.5 Flash Image

Nano Banana Pro

Strongest at photoreal portraits and subject consistency across successive edits of the same face.

ducttape wins on
GPT Image 2 wins on in-image text, multilingual characters, and obeying explicit layout instructions.

ByteDance · Doubao series

Seedream 4

Strongest at stylised illustration, manga/anime aesthetics, and East-Asian design references.

ducttape wins on
GPT Image 2 wins on composition prompts ("three panels, third one vertical") and reference-image fidelity.

Black Forest Labs · Flux family

Flux 1.1 Pro

Strongest at open-weight workflows, ultra-realistic micro-detail, and fine-tuned LoRAs.

ducttape wins on
GPT Image 2 wins on poster and UI-mockup text, multi-line captions, and packaging copy.

Full GPT Image 2 breakdown

Capabilities, API details, release timeline, and head-to-head comparisons with every major competitor.

Read more

The ducttape origin story

How the codename emerged on LMArena, what the duct-tape 1/2/3 variants meant, and how it was confirmed as GPT Image 2.

Read the primer

Frequently asked about GPT Image 2

Plain answers based on confirmed information and LMArena community observations.

What is GPT Image 2?
GPT Image 2 (GPT-Image-2) is OpenAI's next-generation image generation model, the successor to GPT-Image-1. It tested on LMArena under the codename ducttape from late 2025, appearing as duct-tape 1, duct-tape 2, and duct-tape 3 — different variants of the same underlying system.
What is ducttape?
ducttape was the anonymous codename GPT Image 2 used during testing on LMArena. The name appeared in late 2025 and has now been confirmed as OpenAI's GPT Image 2. This site tracked the codename from the beginning.
What's the difference between duct-tape 1, 2, and 3?
On LMArena, GPT Image 2 appeared under multiple suffixes — duct-tape 1, duct-tape 2, and duct-tape 3 — representing different model versions or parameter presets. duct-tape 2 consistently produced the most balanced output, while duct-tape 3 was more aggressive in stylisation.
How is GPT Image 2 different from Nano Banana Pro or Seedream 4?
In blind pairings on LMArena, GPT Image 2 consistently wins on three axes: text-in-image rendering (posters, menus, signage, packaging), layout fidelity when the prompt specifies a structure, and multilingual character rendering for Korean, Japanese, and Chinese. Nano Banana Pro remains stronger for photoreal portraits; Seedream 4 for stylised illustration.
Can I use GPT Image 2 now?
Not yet publicly. The GPT Image 2 API has not been officially released. The only way to experience it currently is through LMArena's blind A/B interface. When the API launches, generation will go live on ducttapeai.app — subscribers hear first.
When will GPT Image 2 launch?
No official date has been announced by OpenAI. The model has been in Arena testing since late 2025 and is expected to launch soon. Subscribe to ducttapeai.app for instant notification the moment it goes live.
Be first to generate

GPT Image 2 generation is coming to this site.

When OpenAI opens the API, generation goes live here. Subscribers hear first — before the tech press. No other emails. Ever.