What's this about?
Images 2.5 improves lighting, texture, identity preservation and multi-turn editing while cutting latency by up to 50%. ChatGPT gains drawing-based Sketch, templates and comments placed directly on an image; the API gains fast Flare and more controlled Sunburst models.
What exactly is “multi-turn consistency”?
It means the model keeps people and untouched details stable while you make several edits. Without it, asking to change a shirt can quietly alter the face, room or hand — like a photo editor that redraws the whole canvas every time.
What's the community saying?
HN 276 points and 369 comments; the launch post had over 18,000 likes.
- Users examined tiny failures in the launch examples — teeth, fingers, colour casts and the recognisable “AI image” aesthetic.
- Some welcomed local comments and Sketch as more useful than ever-longer text prompts.
- Others worried that photorealistic receipts and party photos lower the effort needed for everyday forgery.
- A broader thread expressed fatigue with AI-generated menus, posters and homogeneous commercial design.