ChatGPT Images 2.5 and the Sketch Feature Announced: Simple Drawings Become Art

One of the biggest difficulties when creating images with generative AI models was being able to fully describe the composition in your head using words alone. No matter how detailed the prompt, the exact position of objects in the scene, the point of view or the balance of perspective sometimes ran up against the limits of text. OpenAI has launched ChatGPT Images 2.5, a next-generation image model that removes this communication barrier, together with the revolutionary Sketch feature that comes with it.

Turning even users' roughest doodles into high-resolution artistic output, this update sets a new standard in the AI world with the pinpoint control tools it brings to image editing and the new API models opened up to developers.

Beyond Text: Explain Your Idea by Doodling With @Sketch

The newly announced Sketch capability ends the era of being confined to the prompt box. Users can now type the @Sketch command in the ChatGPT chat area to access a drawing canvas that opens directly within the interface.

Drawings made on this canvas with a mouse, a touchscreen or a digital pen absolutely do not need to look professional. A few stick figures, a roughly drawn horizon line, a simple sun placed in the corner or the rough placement boundaries of objects make a perfect reference for the model.

The system interprets the user's doodle as a compositional skeleton and then, with a short text description added afterwards, turns that draft into a complete scene in a photorealistic or any desired art style. This way, spatial relationships that are hard to convey in text, such as "a figure standing in the left corner but looking to the right", can be communicated to the model flawlessly with a single brushstroke.

Region-Based Comments and Direct Editing on the Image

Another critical innovation OpenAI brought with this launch is the in-context pinning tool, which fundamentally changes the image editing experience.

With previous models, when you wanted to make a small change to a generated image, you had to rewrite the entire prompt or hope the model would change the spot you wanted without breaking the rest of the scene. In the new system, users can click directly on the image to mark a specific region and leave an edit instruction just for that area.

For example, changing only the light source in the background of a portrait, updating the color of a cup on a table or revising a detail of a character's outfit can be done in seconds without disturbing the overall texture of the image or the facial features.

ChatGPT Images 2.5: A New Level in Lighting, Texture and Consistency

The ChatGPT Images 2.5 engine at the center of all these innovations represents a visible leap compared to previous-generation image models:

* Realistic lighting and texture rendering: It produces far more convincing results in natural ambient lighting, shadow transitions and micro-textures such as skin pores and fabric fibers. * Reference preservation: Facial features, object proportions and stylistic elements in photos given to the model as references keep their characteristics when reproduced against different backgrounds or from different angles. * Multi-step editing consistency: When the user requests successive corrections within the same session, instead of forgetting the previous steps and drawing an unrelated image from scratch, the model builds the changes on top of the existing layers in seamless harmony.

Especially for designers and digital agencies working in creative industries, being able to preserve consistency in an image is considered the key step that turns AI from a toy into a professional production tool.

Two New Models for Developers: GPT-Image-2.5 Flare and Sunburst

OpenAI did not limit its new image capabilities to the end-user interface; it also added two powerful models to its API ecosystem:

1. GPT-Image-2.5 Flare

Positioned as the default model for most enterprise and end-user applications, Flare combines high image quality with unprecedented speed. According to data shared by OpenAI, Flare runs with 50 percent lower latency compared to the previous GPT-Image-2 model. It offers an ideal balance for web applications that require instant generation, interactive bots and batches of image variations.

2. GPT-Image-2.5 Sunburst

The Sunburst model was developed for professional workflows that focus on top-level precision and detail accuracy rather than speed. Accepting longer processing times, it renders complex scenes, demanding typographic layouts and millimeter-precise edits at the highest pixel quality.

Who Can Access It, and on Which Platforms?

OpenAI has begun a gradual rollout of ChatGPT Images 2.5 and the Sketch feature. According to the announcement:

* All users of ChatGPT, ChatGPT Work and Codex will be able to benefit from the new image generation capabilities. * The features were launched simultaneously on the web interface, desktop apps (macOS and Windows) and mobile platforms (iOS and Android). * Developers can start integrating both the Flare and Sunburst models into their projects through the OpenAI API dashboard.

This next-generation approach, which can instantly grasp the potential behind a simple doodle, goes beyond text-based commands in human-computer interaction and brings multimodal productivity onto a much more natural footing.