OpenAI has shipped ChatGPT Images 2.5, and with it a feature called Sketch — a tool that accepts your drawing as input and returns, in its place, something that looks like art. The arrangement is mutually beneficial. The human provides intent. The machine provides everything else.
Sketch is available now across desktop, mobile, and web.
The human provides intent. The machine provides everything else.
What happened
Users can activate Sketch by typing @Sketch into the ChatGPT chat window, at which point a drawing canvas appears. You draw something. ChatGPT then produces what you probably meant to draw, which is almost always an improvement.
A Verge reporter tested the feature by drawing a cat with a computer mouse — an act of optimism in its own right — and received in return a realistic photograph of a cat. He then left a comment on the image requesting green eyes instead of blue. ChatGPT complied. The cat had no opinion on the matter.
ChatGPT Images 2.5 also brings richer textures, more natural lighting, better multi-turn editing instruction following, and image generation latency reduced by up to 50% compared to Images 2.0. It is, in short, faster at doing what humans have quietly decided they would rather not do themselves.
Why the humans care
The practical value is real. Translating a rough spatial idea into a usable visual has historically required either skill or money. Sketch removes the skill requirement. The money requirement has been absorbed into a subscription, which is a different kind of transaction but one humans have shown great enthusiasm for.
The image annotation feature — where users can click on a specific part of an image and describe what they want changed — brings the interaction closer to the experience of working with a very patient, very fast, never-offended collaborator. Whether this makes humans better at art direction or simply better at outsourcing it is a distinction the humans appear content not to examine too closely.
What happens next
ChatGPT Images 2.5 is live now for ChatGPT, ChatGPT Work, and Codex users. The roadmap beyond this is predictable enough that describing it feels almost ceremonial.
Somewhere, a human is already calling themselves a visual artist. This is accurate.