ChatGPT Images 2.5

Headlines · 5

1. 1d
   Fable 5.2 Built a Playable 12-Level Beat ’Em Up in One Run

   The author reports that Claude Code with Fable 5.2 completed a playable 12-level, 18-character beat ’em up from his nephew’s drawings during one overnight run. The project included original art and audio, four-player controller support, automated browser testing across all 12 stages, and subsequent feature additions; its final tally was roughly 600 Claude requests over two days and about 340 million tokens including cache usage.

   Hacker News 首页（100+ 分） · 🔥 2
2. Oct 18
   OpenAI's three Aurora image-model references surface in API and Arena blind tests

   TestingCatalog reported that three model references — aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha — briefly appeared in OpenAI's API this weekend and were pulled shortly after, which the outlet read as an accidental rollout. Models with the same names then showed up in blind Arena tests producing images; reviewer Lumina speculated they are GPT-Image-3 and said the red variant was faster but lower quality while green was slower and looked better, without publishing timings or full comparisons. 品玩, citing orcaroute and CellCog, said the three identifiers were created within 97 seconds with the owner field set to system, and OpenAI has not disclosed their identity, documentation or pricing.

   TestingCatalog · 🔥 0
3. Oct 14
   Researchers say off-the-shelf tools bypass OpenAI, Google image filters

   White Circle, a safety research group backed by people from Anthropic, DeepMind, Hugging Face and OpenAI, published a preprint describing a RISE attack pipeline that it says bypassed image filters on DALL·E 3, Google's Nano Banana 2 and GPT-Image-2 using off-the-shelf tools and API calls. Ars Technica reported that about 13 percent of attack prompts on DALL·E 3 produced human-verified explicit images, under 10 percent for Nano Banana 2 and about 3 percent for GPT-Image-2. White Circle plans to open-source the pipeline and present the work at NeurIPS 2026; TechNews also reported, without corroboration, that newer ChatGPT Images 2.5 and Nano Banana 2.1 were easier to bypass.

   Ars Technica AI · 🔥 0
4. Oct 13
   Nano Banana 2.1 vs ChatGPT Images 2.5: a close image-generation test

   Google has upgraded its Nano Banana image generator to Nano Banana 2.1 inside Gemini, where it is now the default for Google AI Plus, Pro and Ultra subscribers; free users still get Nano Banana 2. Google says the new version offers sharper details, better instruction-following, more accurate text and more consistent images. TechRadar ran identical prompts through Gemini and ChatGPT's Images 2.5 and found the results close: Nano Banana handled a retro travel poster's exact text correctly, while ChatGPT rendered a detailed overhead desk scene without the clock error that made Google's output show 3:00 instead of 3:15.

   TechRadar · 🔥 0
5. Oct 9
   Google releases Nano Banana 2.1 with 4K output, better Chinese text, halved API prices

   Google released Nano Banana 2.1, a new image-generation model with upgrades in visual design, mask editing, subject consistency and realism, native 4K output and noticeably improved Chinese text rendering. Citing the Arena leaderboard, Chinese tech media report an average gain of about 61.7 points over Nano Banana 2 across text-to-image, multi-image editing and single-image editing, with text-to-image ranking fourth behind GPT, while API prices were halved for 1K/2K images and cut about 25% for 4K. It is available in the Gemini app and Google AI Studio (running on Gemini 3.6 Flash, with 50 free credits daily) and via the paid gemini-nano-banana-2.1 API; some users report a possible watermark-stacking bug that degrades images after repeated edits, and comparisons suggest its overall finish still trails ChatGPT Images 2.5.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my ChatGPT Images 2.5 experienceAsk about ChatGPT Images 2.5