ChatGPT Images 2.0 rolls out a batch of upgrades aimed at making image generation faster and more flexible. The interface now displays mode options—Classic and Image—so you can toggle between text-only and mixed media workflows. You’ll also see three aspect-ratio presets (Horizontal, Square, Vertical) laid out as big, clickable buttons. That change alone speeds up the setup process: no more manual prompts for dimensions. Behind the scenes, the model’s training data grew by tens of millions of labeled images. OpenAI says that translates into sharper detail, more accurate color reproduction and fewer “weird artifacts” around hands and facial features. Early tests showed generation times dropping from about seven seconds per image to roughly three. That makes it viable for rapid iterations when you’re brainstorming visuals. On the UI side, the update spans four pages worth of options. Page 1 covers basics—start a session, add or remove text prompts. Page 2 lets you adjust creative controls like style strength (0–100 scale) and seed settings for repeatable outputs. Page 3 introduces “reference injection,” where you upload a sample image and the model blends its style with your prompt. Page 4 holds export tools: download in PNG or WebP, or send directly to a ChatGPT thread. Linking this feature into ChatGPT conversations felt central to OpenAI’s plan. Instead of bouncing back and forth between apps, you generate a sketch, flag it for feedback and iterate—all without leaving the chat window. If you click “Try in ChatGPT,” you land in a live demo that shows how text prompts and image edits coexist. It’s a clear push to make visual creation part of everyday dialogue, not a separate design tool.