ChatGPT Image (GPT Image 2)

OpenAI’s GPT Image 2 — the image-generation model behind ChatGPT’s image creation surface, codename “duct-tape” internally. The model leans on three capabilities: pixel-perfect text rendering (including multilingual), cross-image character consistency, and commercial-grade illustration quality. This topic catalogs the prompt libraries, web galleries, and curated resources that have grown up around it — the practical layer that turns “I want an X” into a finished image.

For the Codex CLI / ChatGPT OAuth path that lets Claude Code drive GPT Image 2 generations against the ChatGPT image quota (zero-incremental-cost on Plus / Pro), see Codex CLI’s built-in image_generation tool (ChatGPT OAuth path — zero-incremental-cost on ChatGPT Plus / Pro, drawing from the separate ChatGPT image quota; reachable via codex exec rather than the Platform Images API) in the Claude AI topic. For prompt patterns that transfer cleanly across image and chat models, see Prompt Engineering.

Articles

• ChatGPT Images 2 (GPT Image 2) — Launch Coverage (May 2026) — OpenAI’s official launch (Sam Altman + Gabe / Kuan / Kenji / Alex) plus four creator videos covering the LM Arena leap (250+ ELO jump, 1270 → 1512, instant #1 above Nano Banana 2 / Gemini 3.1 Flash Image Preview), head-to-head testing, beginners tutorial, and 40+ use cases. Headlines: thinking-level image generation that researches + searches the web; multi-image batch (full magazines, manga with recurring characters, room-by-room renovation plans); 2K native, 4K via API; multilingual text rendering; granular select-region editing; the photorealism prompt-keyword unlock for realistic output. Available in ChatGPT + API + Higgsfield + Codex CLI’s image_generation tool.
• awesome-gpt-image-2 — World’s Largest GPT Image 2 Prompt Library (YouMind-OpenLab) — 4.5k-star CC BY 4.0 GitHub repo with 4,400+ prompts in 16 languages, updated daily. Three category groups (Use Cases / Styles / Subjects); three consumption surfaces (GitHub README linear list, YouMind web gallery at youmind.com/gpt-image-2-prompts with masonry + full-text search + AI one-click generation, Raycast integration with dynamic argument substitution). Featured prompts demonstrate the model’s flagship capabilities — exploded product diagrams with multilingual callouts, illustrated city food maps, anime martial-arts scenes with character consistency, live-stream UI mockups with realistic overlays. Prompts mostly curated from public Twitter / 𝕏. Sister to YouMind’s commercial platform.
• Depikt — Curated GPT Image 2 Prompt Library + Generator + Gallery + Critique — Web app at depikt.app with 350+ curated prompts. Smaller and more editorial than awesome-gpt-image-2. Filter by category — Posters / Infographics / UI Mockups / Social Posts / Cinematic / Storyboards / Interior-Food-Fashion / Visual Summaries / Image Edits / Open-Ended Creative. Click “View prompt” to reveal full prompt text under each preview. Generator crafts custom prompts from rough ideas; Critique offers feedback; Gallery surfaces user outputs. Pricing / sign-up / maker info not disclosed on the public landing.

Adjacent coverage in other topics

• Codex CLI’s built-in image_generation tool (ChatGPT OAuth path — zero-incremental-cost on ChatGPT Plus / Pro, drawing from the separate ChatGPT image quota; reachable via codex exec rather than the Platform Images API) — how to actually run GPT Image 2 generations from Claude Code via the Codex image_generation built-in tool, billed against the ChatGPT image quota instead of a Platform API key.
• Higgsfield Overview — Higgsfield’s product-photoshoot path uses GPT Image 2 (via FAL openai/gpt-image-2) under the hood as one of its model backends.
• Higgsfield + Claude Code Ad Agency Workflow — applied GPT-Image-2 + Higgsfield workflow for DTC brand ad creative.
• AI Web Design — adjacent visual-generation surface (web/landing-page focus); shares the spec-prompt vs vibe-prompt taxonomy that applies to GPT Image 2 prompting too.
• OpenAI GPT-5 Prompting Guide — cross-vendor prompting reference. Eagerness control / contradictory-prompt warnings / metaprompting transfer to image-prompt construction even though that guide is text-model-focused.
• Mega Prompt Chest — applied prompt library that overlaps in spirit (community-curated, mixed quality, browse-and-pick).