ChatGPT Images 2.5 Is Here: Faster Generation, Better Editing and a New Sketch Tool

OpenAI has released ChatGPT Images 2.5, a major update to the image-generation technology built into ChatGPT.

The new version focuses on several areas that matter most when creating and editing images: higher image quality, faster generation, better preservation of existing details, and more precise edits.

OpenAI says Images 2.5 can generate images up to 50% faster than Images 2.0, while also doing a better job of maintaining people and other important elements when an existing image is used as a reference.

The update also introduces Sketch, new ready-to-use image templates, improved local editing, easier prompt sharing, and two new image-generation models for developers.

Here’s everything you need to know about ChatGPT Images 2.5.

More Detailed Images With Faster Generation

One of the main improvements in ChatGPT Images 2.5 is overall image quality.

OpenAI says the updated model produces more detailed textures and more natural-looking lighting, helping generated images appear more realistic and refined.

Reference images should also work more reliably, particularly when they contain people.

For example, you could upload a portrait and ask ChatGPT to place the person in a completely different environment or change the visual style. Images 2.5 is designed to preserve the person’s appearance more faithfully instead of unintentionally changing their facial features or other identifying characteristics.

Speed is another significant upgrade.

According to OpenAI, Images 2.5 can reduce generation latency by up to 50% compared with Images 2.0. That makes the process of experimenting with multiple prompts and variations considerably more convenient.

More Precise Image Editing

The editing improvements may be even more useful than the increase in generation speed.

With previous image-generation models, asking for a small change could sometimes cause unexpected modifications elsewhere in the image.

For instance, you might ask the model to replace a single object, only to discover that the character, background, lighting, or composition also changed.

Images 2.5 is designed to reduce this problem.

The model can make more targeted changes while preserving the rest of the image. You can ask it to modify a specific object, background element, or piece of text without unnecessarily regenerating unrelated parts of the scene.

This is especially useful when you’re working on an image through multiple rounds of edits.

According to OpenAI, Images 2.5 also does a better job of preserving previous changes throughout a conversation. As a result, an image should remain more consistent instead of gradually losing details after several consecutive modifications.

Sketch Lets You Draw Your Idea Directly in ChatGPT

OpenAI is also introducing Sketch, a feature designed for situations where describing an image with words isn’t enough.

Instead of writing a detailed prompt to explain exactly where everything should appear, you can draw a quick sketch directly in ChatGPT and use it as a visual reference.

For example, you could sketch:

• The layout of a room
• The position of furniture
• The shape of an outfit
• The arrangement of objects
• The basic composition of a scene
• Where specific elements should appear

The sketch doesn’t need to look polished. Its purpose is simply to communicate your idea visually.

ChatGPT can then use that drawing as a guide when generating the final image.

How to Access Sketch

To use Sketch in ChatGPT:

1. Type @ in the ChatGPT message box.
2. Select the Draw tool.
3. Create a quick sketch showing the composition or element you want.
4. Add your instructions and generate the image.

This could be particularly useful for users who know exactly what they want an image to look like but struggle to describe the composition accurately with text.

Point Directly to What You Want to Change

Images 2.5 also makes localized editing easier by allowing you to indicate where a change should happen.

Instead of describing an object indirectly, you can point out the specific area of the image and explain what you want to modify.

This solves a common problem with complex images.

Imagine a photo containing several people, multiple objects, and a detailed background. Saying “change the chair” may not always be enough for an image model to understand which chair you’re referring to.

With the new editing workflow, you can identify the relevant area directly and then provide your instruction.

Combined with Images 2.5’s improved ability to preserve unrelated parts of an image, this should make small corrections considerably easier.

New Templates Make Image Creation Easier

OpenAI is also adding ready-made templates for common image-creation tasks.

Instead of starting from a completely blank prompt, you can choose a template and customize it with your own content.

Examples include templates for creating:

• Posters
• Product visuals
• Clothing designs
• Promotional graphics
• Other common creative formats

You can then add your own text, images, and instructions to adapt the template to your needs.

The templates are available through the Images section of ChatGPT.

OpenAI notes that these templates are not currently available in ChatGPT Work.

You Can Share an Image Along With Its Prompt

OpenAI is also making it easier to share the ideas behind AI-generated images.

When you share an image created with ChatGPT Images, you can now choose to include the prompt used to generate it.

This gives the recipient more than just the finished picture. They can see the instructions behind the creation and use them as a starting point for their own version.

They can then adapt the prompt using their own images, instructions, or creative ideas.

For creators who frequently share AI-generated artwork or visual concepts, this makes the process more collaborative.

Two New Image Models for Developers

The update isn’t limited to ChatGPT users.

OpenAI has also introduced two new image-generation models through its API:

GPT-Image-2.5 Flare

GPT-Image-2.5 Flare is positioned as the general-purpose option.

It incorporates the improvements introduced with Images 2.5 while offering significantly lower latency. OpenAI says it can deliver up to 50% lower latency than GPT-Image-2.

It’s designed for applications where speed and image quality both matter.

GPT-Image-2.5 Sunburst

GPT-Image-2.5 Sunburst takes a different approach.

It is slower than Flare but places greater emphasis on image quality and precision. OpenAI recommends it for use cases where fine details are particularly important.

Potential applications include:

• Advertising graphics
• Product photography
• Detailed commercial visuals
• High-quality marketing content
• Images containing complex visual elements

Both models are available through the OpenAI API.

ChatGPT Images 2.5 Is Rolling Out Now

ChatGPT Images 2.5 is being rolled out across ChatGPT plans, as well as ChatGPT Work and Codex.

The updated image-generation experience is available across:

• Web
• Mobile apps
• Desktop apps

OpenAI says the existing image-generation limits remain unchanged with the new release.

The company also revealed the scale of its image-generation platform: more than 3 billion images are now created every week through ChatGPT Images and the GPT-Image models available through its API.

That figure shows just how quickly AI-generated imagery has moved from an experimental feature into a mainstream part of the ChatGPT experience.

ChatGPT Images 2.5 vs. Images 2.0

The biggest changes can be summarized as follows:

The result is less about introducing a completely different way of generating images and more about making the entire workflow faster, more controllable, and easier to iterate.

Final Thoughts

ChatGPT Images 2.5 is a meaningful upgrade for anyone who regularly creates or edits images with ChatGPT.

The 50% reduction in generation latency is useful, but the more important improvement may be the model’s ability to make targeted edits without unnecessarily changing everything else.

Features such as Sketch, direct editing annotations, reusable templates, and prompt sharing also make image creation feel less dependent on writing the perfect prompt.

For developers, OpenAI’s new GPT-Image-2.5 Flare and Sunburst models extend those improvements to applications built through the API, with Flare focused on speed and Sunburst targeting more demanding visual tasks.

With billions of images already being generated every week, improvements like these could make AI image creation increasingly practical for everything from casual projects to professional design and commercial content.