ChatGPT Images 2: Why OpenAI Built a New Image Model After Killing Sora

OpenAI announced on Tuesday the release of ChatGPT Images 2, a next-generation image generation model integrated directly into ChatGPT. The model is designed for working professionals to create 'on-brand visuals' within the chat interface, allowing for faster iteration than standalone tools. The launch follows a period of uncertainty around OpenAI's video-generation product Sora, which had been criticised for image capabilities not meeting professional standards, leading to this specialized replacement.

According to OpenAI, ChatGPT Images 2 includes built-in restrictions against generating harmful stereotypes, hate symbols, or depictions of famous living individuals — a change prompted by legal challenges from artists and media companies over copyright and image likeness. CNET reports the output style is 'painterly' rather than photorealistic, and enterprise users can implement further brand-specific guidelines via policy controls. The model is available to ChatGPT subscribers, with additional licensing options for companies.

The article notes this release represents a step back from OpenAI's previous focus on video generation. Sora, unveiled earlier in 2024, promised cinematic-quality output but proved slower and more hallucination-prone in production. Industry analysts quoted by CNET frame this as OpenAI prioritising reliability and commercial applicability over moonshot capabilities. The model's success will partly depend on adoption by design teams needing consistent output at scale.