ChatGPT Images gets GPT‑Image‑1.5: faster, more precise, and easier to create

OpenAI released a new ChatGPT Images experience powered by GPT‑Image‑1.5, now available in ChatGPT and the API. It delivers more precise edits, stronger instruction following, better text rendering, and up to 4× faster generation, plus a new sidebar creation space with presets and likeness reuse. The API adds improved brand/logo preservation and 20% lower image I/O costs, with OpenAI noting clear gains alongside remaining limitations.

Key Points

• New model and experience: ChatGPT Images now runs on GPT‑Image‑1.5, rolling out in ChatGPT for all users and in the API.
• Precision and preservation: Stronger instruction following and precise edits maintain lighting, composition, and likeness across iterative changes.
• Creative power and text: Expanded creative transformations and improved dense/small text rendering produce more natural, faithful results.
• Speed and workflow: Image generation is up to 4× faster with concurrent creation, plus a new sidebar creation space with presets and likeness upload.
• API upgrades and cost: GPT Image 1.5 offers better brand/logo consistency and is 20% cheaper per image I/O than GPT Image 1, supporting marketing and ecommerce use cases.

Sentiment

Mixed-positive. The community broadly acknowledges the technical improvements and benchmark gains, but significant skepticism about benchmark gaming, latency disadvantages, and unresolved ethical concerns around copyright and fake media temper enthusiasm.

In Agreement

• GPT Image 1.5 shows measurable benchmark improvement, doubling scores on localized editing tests compared to the previous model
• The model demonstrates improved prompt adherence and better handling of complex editing tasks without affecting overall image aesthetics
• The new API pricing (20% cheaper than GPT Image 1) is seen as a practical improvement that enables more iteration on the same budget
• Side-by-side comparisons and benchmark results support the claim of meaningful quality improvements in image editing

Opposed

• Latency remains a significant disadvantage, with generation times around 25 seconds compared to competing services at roughly 10 seconds
• Skepticism about benchmark integrity: models released after a benchmark's publication may be specifically optimized for its known test cases, undermining trust in results
• Copyright and training data sourcing remain unresolved ethical concerns, with artists' rights still not adequately addressed
• Improved image generation capabilities accelerate an authenticity crisis, with fake media becoming increasingly convincing and harder to detect
• API availability during launch was inconsistent, with staggered rollout causing confusion despite official announcements
• Economic impacts on creative professions remain a concern, with broader AI image generation potentially devaluing illustrators and designers