Open AI Releases DALL-E 3 Image Editing! (PLUS Free Alternative)

OpenAI has rolled out image editing for DALL·E 3 inside ChatGPT—letting users select regions of a generated image and use natural-language prompts to change them. The feature is available across ChatGPT on the web and on iOS and Android, and it appears to work like an inpainting tool: highlight an area, then ask for edits such as “add bows,” “remove butterfly,” or “turn him into a wizard.” In practice, it can produce clean, localized changes (like swapping clothing details or fixing hands) while preserving much of the surrounding composition.

The rollout matters because it shifts DALL·E 3 from “generate and hope” toward an iterative workflow inside a widely used interface. Users can branch off from earlier results—saving multiple versions as they refine details—without needing separate editing software. A quick demo shows the edit button appearing when a user clicks into a DALL·E 3 image, followed by a selection/highlight workflow similar to painting tools. Prompts can also drive multi-step refinement: for example, changing a top hat reminiscent of Abraham Lincoln, removing an unwanted butterfly, or adjusting a cloak and hat after an initial “wizard” transformation.

Still, the editing has clear limits. Text handling is inconsistent: attempts to correct or insert specific text often fail, with the system sometimes removing text entirely rather than updating it. When a complex layout included a “MattVidPro” label, edits intended to fix the typography resulted in missing or incorrect text, leading to the conclusion that reliable text generation/editing may require other tools. The transcript repeatedly contrasts this weakness with idiogram AI, which is recommended specifically for text.

The tool also struggles with strict consistency. Edits can become “mish-mashed” when prompts require large, style-sensitive changes across broad regions—such as placing a subject on the moon or adding multiple elements that must match the original art style. In one test, the system produced a dramatic scene change with a monster behind a frog; in another, it attempted to add a wizard look but required follow-up corrections to improve the hat and cloak.

Another practical takeaway: the best results come from getting close in the initial generation prompt, then using edits for targeted fixes. The transcript suggests the feature is less about endless fine-tuning and more about correcting specific errors after the first pass.

Beyond OpenAI’s offering, the transcript points to an open-source alternative built with Gradio and distributed via Pinocchio, marketed as BrushNet. The approach runs locally and supports similar “segment and edit” workflows, including swapping the segmented subject into new categories like Lego pieces, chocolate, garlic, or pickles.

Finally, accessibility changes are noted: ChatGPT can be used without an account in some cases, though chat history behavior depends on login. Overall, DALL·E 3 editing in ChatGPT looks like a meaningful step toward democratized image refinement—strong for localized edits and detail fixes, weaker for precise text control and style-perfect consistency.