ChatGPT Images 2.0 supports multi-panel comics in HN tests

• OpenAI's ChatGPT Images 2.0 announcement says the new model handles denser text, stronger instruction following, and multi-image generations, while the launch page summary ties those gains to a new thinking mode.
• In the HN discussion digest, commenter minimaxir immediately pointed builders to the gpt-image-2 model card and pricing docs, which is a good sign that this shipped as an API story, not just a demo reel.
• According to the main HN thread, one of the earliest creator-side tests was multi-panel comics, where commenters said the model kept recurring characters and panel continuity intact.
• The HN discussion digest also surfaced dense text and layout-heavy prompting as practical stress tests, while the system card says thinking mode can use reasoning, tool use, and live web search during image generation.
• Attribution anxiety showed up fast: the main HN thread includes a C2PA discussion around realistic outputs, while OpenAI's developer community launch post published token pricing for gpt-image-2 on day one.

You can read OpenAI's announcement, skim the model page, and check the prompting guide, which already pitches gpt-image-2 for text-heavy images, photorealism, and production editing. The developer community post adds API and Codex availability plus token pricing. Meanwhile, the HN thread quickly narrowed in on the real creator questions: comics, layout fidelity, and whether realistic outputs make provenance metadata more important.

HN got specific faster than most launch coverage. In the HN discussion digest, commenter vunderba said gpt-image-2 handled custom multi-panel comics well and stayed faithful to continuity across panels.

That matters because multi-panel comics usually break on two things at once: recurring character identity and spatial layout. The official announcement also leans into that use case, with OpenAI saying the model can generate multiple distinct images from one prompt and even whole manga-style sequences in a single run, per the launch post.

OpenAI's own framing is that Images 2.0 improved world knowledge, instruction following, and dense text rendering, according to the launch page summary. The system card adds a more interesting detail: thinking mode can use reasoning, tool use, and live web search to turn a basic prompt into a more fully specified final image.

The practical read from HN was less about aesthetics and more about prompt obedience. In the HN discussion digest, commenters framed the model against Google's image stack on prompt adherence and visual fidelity, then stress-tested it with comics and text-heavy layouts instead of generic beauty shots.

The builder-facing part shipped in plain sight. The HN discussion digest notes that minimaxir pointed readers straight to the API model card and pricing docs, and OpenAI's developer community announcement says gpt-image-2 is available in both the API and Codex.

That same post lists token pricing at $8 per 1M image input tokens, $2 per 1M cached image input tokens, $30 per 1M image output tokens, $5 per 1M text input tokens, $1.25 per 1M cached text input tokens, and $10 per 1M text output tokens. OpenAI's prompting guide calls gpt-image-2 the default pick for new builds, especially for text-heavy images, photorealism, compositing, identity-sensitive edits, and production workflows where fewer retries matter.

One of the more grounded HN side conversations had nothing to do with image quality. In the main HN thread, commenter madrox raised C2PA as a way to affirm image sources and argued that unlabeled realistic images may increasingly be treated with suspicion.

OpenAI's system card says Images 2.0 keeps the earlier safety stack and adds safeguards for the new reasoning and tool-use behavior. That does not resolve the provenance debate, but it does show how quickly creator talk around this release split into two lanes: better comics and layout control on one side, and source labeling for realistic work on the other.