OpenAI has not officially announced GPT Image 2. But the model is already out in the wild. Three codenamed variants — maskingtape, gaffertape, and packingtape — have appeared in anonymous Chatbot Arena testing over the past few weeks, and ChatGPT users are reporting that they are randomly triggering the new model during regular image generation sessions. The AI image generation space is about to get a major shakeup.
What We Know From the Leaks
Developers first noticed backend changes in ChatGPT’s image pipeline in late 2025. Rendering shifted from a two-stage process to a single inferential pass, output fidelity improved noticeably in skin textures and shadow gradients, and metadata tags appeared in generated PNG files that did not exist in GPT Image 1.5 outputs.
Then came the Arena testing. The three codenamed models showed significant improvements in spatial reasoning, multi-object composition, and stylistic range. Users who encountered them in blind comparisons consistently rated them higher than GPT Image 1.5 — and in several categories, competitive with Google’s Nano Banana Pro.
The timing makes sense too. OpenAI shut down Sora on March 24, 2026, after the video generation tool burned through an estimated $15 million per day in compute costs while generating only $2.1 million in total revenue. That freed up an enormous amount of GPU capacity, which analysts believe was redirected toward GPT Image 2 development.
Why This Matters
GPT Image 1.5 was already a significant step forward — four times faster than its predecessor, with much better prompt adherence and text rendering. But it still trailed Nano Banana Pro in raw photorealism. If the Arena results hold up, GPT Image 2 could close that gap while pushing ahead in areas where GPT-based models already lead: speed, creative flexibility, and instruction following.
For anyone who creates visual content — marketers, designers, content creators, indie game developers — this means the best available tools are about to get dramatically better, and likely cheaper as competition between OpenAI and Google intensifies.
How to Prepare Now
You do not need to wait for the official release to start building GPT image generation into your workflow. The underlying technology is already accessible, and the skills you develop with current tools will transfer directly to GPT Image 2 when it launches.
Start experimenting with prompt techniques: specify lighting, composition, art style, and mood in your descriptions. Practice iterative workflows where you generate multiple variants and refine through follow-up instructions. These habits will give you a significant head start when the next-generation model drops.
If you want to try GPT Image 2 for fast, high-resolution AI image generation with a simple interface, it is a great place to start — no API keys, no complex setup.
Frequently Asked Questions
What is GPT Image 2? GPT Image 2 is OpenAI’s upcoming next-generation image generation model, expected to succeed GPT Image 1.5. It has not been officially released yet but has appeared in anonymous Arena testing under the codenames maskingtape, gaffertape, and packingtape.
When will GPT Image 2 be released? OpenAI has not confirmed a release date. Based on the Arena testing timeline and the compute freed up by Sora’s shutdown in March 2026, industry analysts expect an announcement between April and June 2026.
How is GPT Image 2 different from GPT Image 1.5? Early testing shows improvements in spatial reasoning, compositional accuracy, photorealism, and stylistic range. It is expected to be integrated into the GPT-5 model family rather than operating as a standalone image model.
Where can I try GPT Image 2? You can try GPT Image 2 at gptimage2ai.com, which offers AI-powered image generation through a simple interface without requiring API setup or a ChatGPT subscription.
Is GPT Image 2 better than Nano Banana Pro? Based on limited Arena testing, GPT Image 2 appears to close the photorealism gap with Nano Banana Pro while maintaining advantages in speed and prompt adherence. A definitive comparison will only be possible after the full public release
