OpenAI ships GPT Image 2.5 with faster generation, sketch canvas and precise inpainting
卡尔的AI沃茨 · wechat · 2026-09-09
OpenAI unexpectedly released GPT Image 2.5 ahead of DevDay, focusing on faster generation, higher resolution, and consistency across multi-turn edits, plus new canvas sketching and point-annotation precise editing features.
Two API variants shipped alongside: GPT-Image-2.5 Flare (50% faster at same quality) and GPT-Image-2.5 Sunburst (higher precision, slower). In the author's 150-generation test with a blue cube, 2.5 shows far less deformation, removing the need for structured prompts with fixed camera and lighting constraints.
The new drawing features are web-only for now (unavailable in codex). Fifteen templates cover posters, logos, infographics, product shots and more. Side-by-side comparisons show clear gains in logo and character consistency—one annotated mole appeared in a single attempt—though stylization is softer than image2. The author also built a 12-frames-per-second stop-motion clip to verify shape stability.
The real shift is interaction: sketching replaces writing, clicking replaces describing, and better consistency removes stability boilerplate—good prompts are no longer a prerequisite for AI image generation.
Related event: OpenAI launches ChatGPT Images 2.5 with Flare and Sunburst API models(43 posts)→
More from coding & agent
- Cursor launches Agent SDK: agents as code with Markdown prompts and TypeScript tools — soleio · 2026-10-27
- Vibecheck, the 'usable formal methods' hackathon, sells out two floors in SF — spikedoanz · 2026-10-27
- How do you verify AI agent outputs beyond unit tests and LLM-as-judge? — Practical_Green5775 · 2026-10-27
- AI agents may soon become the only way to run code, not just write it — IgorCarron · 2026-10-27
- Hermes Agent used as browser QA tester finds 8 issues in a bug-seeded site — techlatest_net · 2026-10-27
- Open-source FableCut lets AI agents edit a real video timeline via remote MCP — FastPresence9799 · 2026-10-27