OpenAI has launched ChatGPT Images 2.5, a significant upgrade to its image-generation tool that delivers faster performance, better detail preservation, and new creative features like a Sketch tool for drawing directly into the platform. The update, announced on September 8, 2026, introduces two specialized model variants and addresses a key frustration for creative professionals: maintaining image consistency across multiple edits within a single conversation.

What Makes ChatGPT Images 2.5 Different From Previous Versions?

The new version represents a meaningful leap forward from ChatGPT Images 2.0, which debuted in April 2026. OpenAI focused on three core improvements: speed, consistency, and creative control. The most dramatic performance gain is the 50% faster generation time, which means users can iterate on their ideas more quickly without waiting for renders.

Beyond speed, the model now generates images with "more natural lighting and richer textures," according to OpenAI's announcement. Perhaps more importantly for professional workflows, ChatGPT Images 2.5 preserves distinctive details like facial features and identifying characteristics across multiple generations and edits within a single conversation, a capability that was unreliable in earlier versions.

How Does the New Sketch Feature Work?

The Sketch tool is the headline addition to this release. Users can now draw directly into ChatGPT using their finger or stylus on a touchscreen, and the model will use that sketch as a visual reference point to generate a polished, finished image. This bridges the gap between rough ideation and final output, letting creative professionals sketch a layout, visual direction, or composition, and have the AI build on that foundation.

The feature is particularly useful for those with a stylus like an Apple Pencil, though touchscreen drawing is supported across devices. This addition signals OpenAI's recognition that many users want more control over the creative direction before the AI takes over, rather than relying entirely on text prompts.

Key Features and Technical Improvements

• Dual Model Variants: GPT-Image-2.5 Flare prioritizes speed, generating high-quality images two to four times faster than the previous model, while GPT-Image-2.5 Sunburst sacrifices speed for greater precision and pixel-perfect detail consistency, ideal for product photography and marketing materials.
• Iterative Editing Stability: The model now preserves details users have not asked to change, even in complex scenes with multiple subjects and busy backgrounds, ensuring that earlier edits remain stable as users request successive modifications.
• Enhanced Reference Photo Handling: Distinctive facial features and identifying details are now carried through more reliably across multiple image generations and edits, addressing a major limitation of version 2.0.
• Inline Comments: Users can now add targeted comments directly on images to specify exactly what they want changed, reducing ambiguity in edit requests.
• Template Library: New templates for common formats like flyers and product photos help users get started faster with professionally structured layouts.
• Prompt Sharing: Users can share their prompts alongside generated images, enabling colleagues and friends to iterate on ideas collaboratively.

How to Use ChatGPT Images 2.5 for Professional Creative Work

• Start With Sketch: Use the Sketch tool to rough out your composition, visual direction, or layout before letting the AI generate the final image, ensuring your creative intent is preserved from the start.
• Choose Your Model Variant: Select GPT-Image-2.5 Flare for quick iterations and brand assets, or GPT-Image-2.5 Sunburst when you need pixel-perfect consistency for product photography and marketing materials that must be perfectly on-brand.
• Leverage Inline Comments: Use the new inline comment feature to specify targeted edits on specific areas of your image, reducing the need for vague text instructions and improving edit accuracy across multiple rounds.
• Build on Previous Edits: Make successive edits with confidence that earlier changes will remain stable; the model now maintains consistency across long conversations with multiple edit requests.
• Share and Collaborate: Export your prompts alongside images so team members can understand your creative decisions and iterate on your work without starting from scratch.

OpenAI has also built in security guardrails to prevent harmful content generation. The platform uses C2PA metadata and invisible watermarking techniques to ensure generated images can be identified as AI-created, helping prevent unauthorized reuse and deepfakes.

Where Can You Access ChatGPT Images 2.5?

The new model is available immediately in ChatGPT, ChatGPT Work, and Codex across desktop, mobile, and web platforms. Developers can also access both GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst through separate application programming interfaces (APIs), allowing integration into third-party applications and workflows.

The release reflects OpenAI's broader strategy of making its image-generation capabilities competitive with specialized tools while maintaining the convenience of an integrated chat interface. By addressing the specific pain points of creative professionals, such as detail preservation and iterative editing stability, ChatGPT Images 2.5 positions itself as a viable alternative to standalone image-generation platforms for many workflows.