The guide · breakdown
GPT Image 2 — the full leak.
OpenAI's next image model has been surfacing under a string of codenames in the LMArena battle rotation — near-photorealistic faces, text that actually renders, code and formulas with zero distortion. Here's every codename spotted so far, how to hit the model yourself right now, what makes it a genuine leap over 4o, and a few test prompts to prove it.
The full leak
🔒 Drop your email to reveal the full leak
Get the full leak, free.
Drop your email and every codename, the access method, and the test prompts unlock instantly, plus every other guide and resource in the library. One email a week, unsubscribe anytime.
All known codenames (in order)
| Codename | When spotted | Notes |
|---|---|---|
hazel-gen | December 9, 2025 | First sighting on LMArena |
hazel | December 2025 | Same batch |
chestnut | December 2025 | Believed to be Image-2-mini (lightweight) |
hazelnut | December 2025 | Believed to be Image-2 flagship |
maskingtape-alpha | April 2026 | Latest batch — all pulled from Arena |
gaffertape-alpha | April 2026 | Same batch |
packingtape-alpha | April 2026 | Same batch |
There might be new codenames coming out, so hurry before it gets changed.
How to access it right now
What makes it different
- Not based on GPT-4o — a completely separate architecture.
- Near-photorealistic faces with accurate skin texture and lighting.
- Text inside images actually renders correctly (4o fails at this).
- Code, formulas, and labels with zero distortion.
- The biggest leap from OpenAI in image quality since DALL-E 3.
Test prompts to try
Cinematic portrait:
Text rendering:
Code on a page:
Candid street photography:
Original guide by Artem Novitckii. More from him on Instagram, in the Skool community, and across his resource library.