Codex Can Now Draw: OpenAI Unveils ChatGPT Images 2.0

OpenAI announced the official release of ChatGPT Images 2.0, positioning it as a state‑of‑the‑art image model capable of handling complex visual tasks and producing ready‑to‑use graphics.

“From today, all ChatGPT and Codex users can use it.”

Codex (v0.122.0, model gpt‑5.4) can now generate images directly. The author prompted the model with “Draw a diagram: what should a SaaS architecture look like?” and, without any image‑model API key or extra plugins, received a layered SaaS architecture diagram saved to .codex/generated_images/ in under a minute.

The generated diagram shows clear layering (user, access, multi‑tenant application, platform capabilities, data), a dedicated “observability & security” column, design‑principle sticky notes, and legend lines distinguishing synchronous requests, asynchronous streams, and integration.

Clear layering : user, access, application, platform, data.

Support capabilities isolated : a vertical “observability & security” area covering monitoring, alerts, tracing, audit, key management.

Design principles as sticky notes : multi‑tenant isolation, elastic scaling, high availability, automated ops.

Legend conventions : line styles for sync, async, and integration relationships.

Notably, the Chinese labels render cleanly—previously AI‑generated diagrams often produced garbled characters. Labels such as “multi‑tenant SaaS platform” and “file storage OSS/S3” appear crisp, with mixed Chinese‑English layout well‑aligned.

The workflow requires no image‑model key, no switch to the ChatGPT web UI, and stays within the coding terminal, highlighting the seamless integration of image generation into Codex.

This update gives Codex its first “brush” as a programming agent, enabling it to generate code, text, and now images in a single loop.

Key technical details of Images 2.0:

Resolution : up to 2K.

Aspect‑ratio support : full coverage from 3:1 to 1:3.

Knowledge cutoff : December 2025.

Thinking images : initially available to ChatGPT Plus, Pro, and Business users; enterprise rollout forthcoming.

Compared with DALL‑E 3, Images 2.0 reliably renders text, eliminating the misspellings that plagued earlier models.

OpenAI demonstrated precise instruction execution with a macro‑level rice‑grain close‑up and legible handwritten English notes, showing the model’s ability to handle fine‑grained visual details.

Multilingual rendering is explicitly improved, supporting Japanese, Korean, Chinese, Hindi, Bengali, and other non‑English scripts. A Japanese manga example correctly places dialogue, onomatopoeia, and compound words.

Aspect‑ratio flexibility enables practical use cases: horizontal banners, slide decks, vertical posters, long‑form graphics, and even exaggerated pull‑rope scenes without breaking character proportions.

Images 2.0 delivers consistent quality across photos, cinematic frames, pixel art, and comics, with refined texture, lighting, composition, and detail.

Suggested real‑world applications include game prototype creation, storyboard design, marketing concepts, and media‑specific asset generation, which can accelerate early development stages for game creators and indie developers.

The model can also “understand and draw” complex logical diagrams, exemplified by a correct step‑by‑step illustration of Cantor’s diagonalization proof, integrating textual reasoning with visual output.

Implications for agents: the workflow expands from “generate code + write text” to “generate code + write text + draw images,” closing more loops within a single agent.

Potential developer scenarios:

Frontend development : generate icons, banners, and empty‑state illustrations alongside React components.

Product prototyping : produce UI sketches, flowcharts, and architecture diagrams while writing PRDs.

Documentation & presentations : auto‑create accompanying graphics for technical docs and PPT covers.

Game & asset creation : generate characters, scenes, and UI assets in sync with game logic code.

Codex’s new drawing capability reshapes the “software factory” vision, now covering code, tests, deployment, documentation, and visual assets.

Images 2.0 is publicly available to all ChatGPT and Codex users. “Thinking” image generation is limited to Plus/Pro/Business tiers, with enterprise access pending. Mobile users should update to the latest app version.

The API model ID is gpt-image-2. Pricing is not fully disclosed; OpenAI states costs depend on quality and resolution, and teams are encouraged to run demos.

Overall, the most significant shift is the transition of image generation from a creative “inspiration tool” to a production‑grade asset generator, delivering directly usable graphics that meet precise textual, layout, and multilingual requirements.

For full‑stack, solo developers, and content creators, this new capability may prompt a redesign of existing workflows to incorporate on‑the‑fly visual generation.