ChatGPT rolls out in-chat image editing and sketch-to-image

Anyone who uses AI image generators knows the familiar frustration of the reroll lottery. You get a nearly perfect image, but one hand looks weird or a sign has gibberish text. You ask the model to fix it, and it wipes the entire composition and hands you something completely different.

OpenAI is tackling that exact headache. The company rolled out native image generation and editing inside ChatGPT across web, iOS, and Android for all user tiers, while officially retiring the standalone DALL-E GPT.

Precision editing replaces the reroll lottery

The marquee addition here is the in-chat image editor.

Now, when ChatGPT delivers an image, you can click on it and grab a selection brush. Just highlight the exact area you want modified, type your prompt in the chat box, and let the model update only that section. You can also skip the brush entirely and just describe the regional change in words.

The editor lets you change aspect ratios, drop in specific text, and even make backgrounds transparent. You can also upload your own existing pictures and ask ChatGPT to tweak them.

OpenAI notes, however, that the brush tool is not surgical precision just yet. Edits can sometimes bleed slightly beyond the highlighted boundaries.

From rough sketches to polished art

Mobile users are getting an especially fun workflow. On iOS and Android, typing "@" and choosing Sketch opens up a blank canvas right above the message box.

You can doodle a rough outline of a cabin with your finger, add a prompt asking for a watercolor landscape with a red house surrounded by trees, and watch ChatGPT turn the scribble into finished artwork.

OpenAI also introduced design templates for things like posters and logos. The model asks interactive follow-up questions to nail down details like style, palette, and typography before rendering.

Why it matters

We think this is the kind of mundane friction-killer that actually turns generative AI into daily utility.

For casual creators, non-designers, and marketers, fixing mistakes without destroying the rest of the image saves massive amounts of time and patience. It bridges the gap between typing a wish into a void and actually directing creative output.

Midjourney and open-source diffusion models have had inpainting for a while, but usually tucked behind clunky Discord commands or complex web UIs. OpenAI's move here is brilliant because it folds inpainting and doodling directly into the most popular conversational interface on earth.

It is not foolproof—bleeding beyond selection boundaries and slight stylistic drift remain quirks of diffusion models. But making image iteration this accessible puts serious pressure on dedicated graphic design tools.