Editing ChatGPT images

You can now “edit” your AI images in ChatGPT 4, but I’m not sure if it’s worth your time.

For those who subscribe to ChatGPT, you can generate images with your prompts. Now, within ChatGPT itself, you can select a part of your image with a brush, and then type in an additional prompt to regenerate that section.

For example, you can replace a dog with a cat, or you can add earrings to a person’s portrait. Or you could remove things that you don’t want in the scene.

This “editing” feature has been long available in apps like Midjourney (“Vary Region”), Adobe Photoshop (“Generative Fill”) and Stable Diffusion (“Inpainting”), and I’m glad that ChatGPT’s DALL-E 3 engine has caught up.

However, the feature is unpredictable and 70% of the time, I received very ugly or weird results. Sometimes, ChatGPT wouldn’t even follow the prompt I gave.

Midjourney, on the other hand, offers a much better experience. I guess this is because Midjourney is focused on imaging while ChatGPT is a multimodal platform that tries to excel at  language, data analysis and imaging.

Anyway, do try out the new feature if you are a ChatGPT subscriber, but don’t spend too much time on it if you aren’t getting your desired results!

Original LinkedIn post.