HN Turns ChatGPT Images 2.0 into a Prompt-Adherence Shootout

Why the thread took off

A Hacker News post on ChatGPT Images 2.0 reached 1002 points and 921 comments after OpenAI published the release page on April 21, 2026. The official page frames the model around image generation, precision, control, multilingual text rendering, varied aspect ratios, and richer visual reasoning examples. HN treated that less like a launch note and more like a public benchmark suite.

The community test was prompt adherence

The most useful signal in the thread was how quickly users moved past "looks realistic" into harder checks. Commenters compared the model against Google image models, shared deliberately awkward prompts, tested dense typography, and looked for whether the output could obey multiple constraints at once. Community discussion noted that the new model looks strong on editorial layouts, infographics, multilingual signage, and scenes where text has to stay readable rather than dissolve into decorative noise.

Cost and provenance mattered too

HN also focused on the parts that determine whether people use an image model in production: price per render, predictable quality at lower settings, and how generated images can be identified later. One thread branch pointed to C2PA provenance metadata as a positive signal, while also noting the weakness of any system that can be stripped by bad actors. That made the debate more technical than the usual "AI art" argument.

What this says about image models now

The energy around the post came from a shift in expectations. Image generation is no longer being judged only on photorealism. Developers and designers are asking whether a model can follow instructions like a layout engine: preserve text, keep character identity, fit a requested format, reason over source material, and repeat the result without burning budget. The HN reaction suggests ChatGPT Images 2.0 is being evaluated as a work tool, not as a novelty.

OpenAI release page | Hacker News discussion