ChatGPT Images 2.5 is OpenAI’s latest image-generation model, promising sharper details, more accurate editing, and generation times up to 50% faster than Images 2.0.
According to OpenAI, people now create more than three billion images every week across ChatGPT Images and its GPT-Image models in the API. The company is responding to that demand with improvements aimed at both everyday ChatGPT users and developers building image-generation features into their products.
The new model is rolling out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. OpenAI is also releasing two corresponding API models called GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
Better reference image fidelity and more controlled editing
One of the main improvements in ChatGPT Images 2.5 is how it handles reference images. OpenAI says the model is better at preserving recognisable subjects when placing them in different locations, compositions, or visual styles.
Lighting and textures should appear more natural, while distinctive facial or physical features are more likely to remain consistent with the original reference. This could make a noticeable difference when creating variations of the same person, product, or character, where earlier image models sometimes changed important details between generations.
Precise editing is another major focus. ChatGPT Images 2.5 is designed to change only the requested part of an image while leaving its other elements intact. For example, users should be able to replace a background, modify a product, or update text without unexpectedly altering the subject or overall composition.
This sounds awesome. Generating an impressive first image is useful, but being able to refine it without having unrelated details change is what makes an image-generation tool practical for ongoing creative work.
The model should also maintain better consistency across multiple rounds of editing. Previous changes are more likely to remain in place as new instructions are added, with less degradation in image quality over a longer conversation.
For developers and businesses, this could make it easier to produce a consistent collection of product images, campaign assets, presentation graphics, or interface concepts without rebuilding the image after every requested change.
ChatGPT Images 2.5 can handle more complex instructions
OpenAI says ChatGPT Images 2.5 has improved its understanding of detailed visual instructions. This includes better handling of complex layouts, transparent backgrounds, artistic styles, and images containing real-world information.
The model is also intended to follow a requested composition and visual direction more closely, even as prompts become increasingly specific. This should be helpful when an image needs to follow a particular brand style or maintain the same layout across several variations.
AI-generated artwork continues to raise broader questions about creativity, originality, and how these tools fit alongside human artists. Regardless of where you sit in that debate, better consistency and editing control should make the technology more useful for people who already incorporate it into their workflows.
OpenAI has also reduced image-generation latency by up to 50% compared with Images 2.0. The difference should be particularly noticeable when repeatedly generating variations or making several small edits to reach the desired result.
Sketch lets you draw directly inside ChatGPT
Alongside the new model, OpenAI is introducing Sketch, a feature that lets users draw directly within ChatGPT and use that drawing as a visual reference.
Rather than attempting to describe every element of a composition through text, users can sketch the approximate layout of a room, the shape of an outfit, or the position of objects within an image. ChatGPT can then turn the rough drawing into a completed image while following additional instructions about its content and visual style.
You do not need any serious drawing skills to use it. The sketch primarily acts as a guide, helping ChatGPT understand where elements should be placed and how the final composition should look.
Sketch can be accessed by typing “@Sketch” in ChatGPT.
Templates, image comments, and shareable prompts
ChatGPT is also gaining templates for common creative formats. Users can start with options such as “Poster” or “Merch”, then supply the information, design elements, and style they want included.
These templates should make the process less intimidating for anyone who knows what they want to create but is unsure how to structure an effective image prompt.
Comments can now be placed directly on an image as well. This provides a more targeted way to request changes by indicating exactly which area needs attention, rather than describing its position entirely through text.
Users can also include their original prompt when sharing an image. Other people can reuse that prompt with their own photos and details, making it easier to share a particular visual concept rather than only the finished result.
GPT-Image-2.5 Flare and Sunburst arrive in the API
Developers will have access to two versions of the new model through the Images API.
GPT-Image-2.5 Flare is positioned as the default option for most applications. It carries across the improvements to quality, editing, and speed, with OpenAI claiming higher-quality output than GPT-Image-2 at 50% lower latency. The company suggests it for creator content, social media, product experiences, visual search, rapid prototyping, and high-volume image generation.
GPT-Image-2.5 Sunburst is intended for more detailed creative and editing workflows where precision is more important than generation speed. Potential uses include polished product imagery and production-ready advertising assets.
OpenAI says ChatGPT Images 2.5 continues to use prompt and image checks intended to reduce harmful outputs. Generated images also carry C2PA metadata and invisible watermarking to help identify content created with OpenAI’s tools.
ChatGPT Images 2.5 is rolling out now across all ChatGPT tiers, including ChatGPT Work and Codex, on desktop, mobile, and web. OpenAI did not include API pricing for GPT-Image-2.5 Flare or GPT-Image-2.5 Sunburst in the supplied announcement.
