Three billion images a week. That’s the number OpenAI is now citing for combined usage across ChatGPT Images and its API-based GPT-Image models. So when the company announced Images 2.5, it wasn’t just a quality update. It was OpenAI signaling that image generation is now core infrastructure, not a side feature.
The headline numbers are real. Images 2.5 cuts generation latency by up to 50% compared to Images 2.0. That alone matters for any product team running image generation at scale. But the more interesting story is what OpenAI is doing with editing consistency and reference fidelity, two areas where competitors like Midjourney, Stability AI, and Adobe Firefly have struggled to deliver reliable results in production workflows.
What’s actually new in the model
Images 2.5 is better at preserving subjects from reference photos across different visual styles and compositions. Lighting and textures are more natural, and distinctive features are more likely to carry through when you’re remixing an image. For creative teams, that means fewer regeneration cycles. For developers building reference-led workflows into products, it means results stay anchored to the source instead of drifting.
The editing improvements are also worth taking seriously. The model is now better at changing only what you ask it to change, leaving the rest of the image intact. That sounds basic, but it’s been a consistent pain point across most image models. Multi-turn editing consistency is also improved, meaning earlier changes hold through subsequent edits instead of degrading over time. That’s a meaningful shift for anyone building iterative creative tools.
New consumer features in ChatGPT
On the product side, OpenAI added several features aimed at lowering the barrier to creative work inside ChatGPT:
- Sketch lets you draw directly in ChatGPT as a visual guide. You rough out a layout or shape, add a style description, and the model builds from it. Type “@Sketch” to access it.
- Templates give you starting points for common formats like posters, flyers, and product photos, so you’re not starting from a blank prompt every time.
- Image comments let you annotate specific parts of an image for more focused editing instructions.
- Prompt sharing lets you share not just the output image but the prompt behind it, so others can remix the idea with their own photos and details.
The Sketch feature is the most interesting of the bunch. Drawing as a communication layer between user and model is genuinely useful for spatial or compositional ideas that are hard to describe in text. It also puts OpenAI closer to what tools like Adobe Firefly and Canva have been building toward, though with a more conversational interface.
Two new API models: Flare and Sunburst
For developers, the bigger news is the two new API models. GPT-Image-2.5 Flare is the default choice for most applications. It delivers higher quality than GPT-Image-2 at 50% lower latency and is aimed at creator content, product imagery, social assets, and high-volume generation. GPT-Image-2.5 Sunburst is the premium tier, built for production-ready campaign creative and polished product imagery where tighter control across edits matters more than speed.
That two-tier structure makes sense. Not every use case needs maximum quality at every step. Giving developers a faster, cheaper option alongside a precision-focused one is how you build an image API that works across different product contexts.
Why this matters beyond the feature list
OpenAI is not just competing with Midjourney or Stable Diffusion here. The real competitive pressure is coming from Adobe, which has deep creative workflow integration, and from startups like Ideogram and Recraft that have focused specifically on text rendering and design-quality output. Images 2.5 doesn’t appear to be making a specific claim around text rendering, but the improvements to complex layouts and transparent backgrounds suggest OpenAI is moving in that direction.
Images 2.5 is available now to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. The API models are available to developers today. For teams already building on the Images API, upgrading to Flare is probably the obvious first move. The latency improvement alone justifies the switch.
