OpenAI dropped ChatGPT Images 2.5 on Tuesday, and the headline isn’t just about better pixels. It’s about speed and control. The company claims image generation latency has been slashed by up to 50% compared to version 2.0. For anyone who has stared at a spinning wheel waiting for a render, that cut in wait time is tangible. But the real shift here is how you talk to the model. You no longer need to be a poet with words to get a specific visual result.
Enter Sketch. This new feature lets you draw directly into the chat interface to guide the AI. Type @Sketch, and a canvas appears. I tested it with a mouse-drawn cat that looked more like a potato with ears. The result? A realistic photo that actually resembled a feline. You can then layer text instructions on top of your doodle. It bridges the gap between abstract intent and concrete output, letting users who struggle with descriptive prompts use spatial reasoning instead.
Editing also got a surgical upgrade. Instead of regenerating an entire image because one detail is off, you can now leave comments directly on specific parts of the generated picture. In my test, I flagged the cat’s eyes and requested they be green instead of blue. The model adjusted only that element while preserving the rest of the composition. OpenAI says the underlying model now handles “more natural lighting and richer textures” and follows editing instructions better across multiple turns. This iterative refinement feels less like gambling and more like collaborating.
These updates are live now for ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. While better lighting and texture are welcome, the reduction in friction is the real story. By allowing doodles and pinpoint edits, OpenAI is acknowledging that most people think visually, not verbally. The question now is whether this lowers the barrier enough to make AI image generation a daily utility rather than a novelty. If latency stays low and accuracy holds up, we might finally stop fighting the prompt box.
