OpenAI is rolling out ChatGPT Images 2.5, its newest image generation model, and it brings a solid mix of speed and quality upgrades that should make your AI art sessions a lot less frustrating. People are already generating over 3 billion images every week across ChatGPT and the API, and this update is built to make that process faster and noticeably more accurate.
What makes Images 2.5 different?
The biggest wins here are sharper details, more natural lighting, and textures that finally look believable instead of plastic. The model is also much better at sticking to your reference photos, so if you’re turning a picture of yourself into a new scene or style, your face and features should stay recognizable instead of drifting into someone else entirely.
Editing gets smarter too. Ask for one change, like swapping a background or tweaking a product, and Images 2.5 should leave everything else alone. That consistency holds up across multiple rounds of edits, so your earlier tweaks won’t randomly get wiped out a few prompts later. And if speed is what you care about, generation is now up to 50% quicker than Images 2.0.
Sketch lets you draw your idea before AI finishes it
OpenAI added a feature called Sketch that lets you doodle your idea right inside ChatGPT, whether that’s a room layout or an outfit concept, and have the AI turn it into a polished image. Pair that with new templates for posters, flyers, and product shots, along with the option to leave comments directly on images for more precise edits, and you get a lot more control without needing any design skills.
People can also share the exact prompt behind an image now, so if someone creates something you love, you can recreate it using your own photos and details. On the developer side, OpenAI is also launching GPT-Image-2.5 Flare and Sunburst in the API, bringing the same speed and quality gains to apps and products built on top of it.
Images 2.5 is live now for all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web.
