The command line is becoming a canvas, and now your worst artistic impulses can become production-ready visuals in seconds.
The Summary
- OpenAI launched ChatGPT Images 2.5 with Sketch, letting users draw rough doodles directly in ChatGPT and transform them into polished AI images
- Type @Sketch in the chat box to activate a drawing window, then describe what you want the final image to look like
- Users can now leave comments directly on parts of images, enabling iterative refinement through visual markup
- This bridges the gap between conception and execution for anyone who can point and type but can't draw
The Signal
OpenAI just collapsed the distance between idea and artifact. ChatGPT Images 2.5 introduces Sketch, a feature that turns ChatGPT into a visual translation layer between your brain and the screen. You draw something bad. You tell it what you meant. It renders what you wanted.
The interface is deliberately simple. Type @Sketch, get a drawing window, use your mouse or trackpad. No separate app. No export and upload dance. The friction of "I have an idea" to "here's the image" just dropped to near zero.
"It successfully turned my bad doodle of a cat that I drew with my computer mouse into a realistic photo, per my instructions."
Here's what matters: this isn't about replacing designers. It's about replacing the moment when someone says "I can't draw, but what I mean is..." and everyone squints at a napkin sketch. Now that napkin sketch becomes a brief, and the brief becomes a render. The system accepts sketches and reference photos, treating both as input types for generating "more personalized, polished images."
The new annotation feature is the quiet power move. Users can now comment directly on parts of images. That means revision cycles happen in-context. "Make this darker." "Add a window here." "Remove that shadow." You're not describing changes in abstract text anymore. You're pointing at pixels and issuing instructions like you would to a junior designer sitting next to you.
Key capabilities:
- Sketch-to-image transformation using rough drawings as visual briefs
- In-line image annotations for targeted revisions
- Accepts both sketches and reference photos as source material
- Native integration into ChatGPT interface, no app switching required
This shifts the bottleneck in creative work. The constraint used to be "can you draw it?" Now it's "can you describe what you drew?" Most people can do the second thing. OpenAI is betting that removing artistic skill as a prerequisite for visual creation opens up a much larger market than people who can already use Photoshop or Midjourney with precision.
The Implication
Watch what happens to creative briefs over the next six months. The standard workflow of "write a description, hope the designer understands, wait for revision cycles" is about to get faster and messier. Anyone with an idea and a mouse can now produce a visual reference instead of a paragraph trying to describe one.
For agents building visual content, this is training data for intent. Every sketch paired with a prompt and a final image teaches the model what humans mean when they point and say "like this, but more." That's the foundation for agents that can iterate on design work without human supervision. The person who learns to prompt Sketch effectively today is training the art director agent of tomorrow.