ChatGPT Images 2.5: Sketch Brings Drawing to Chat
OpenAI expands image generation in ChatGPT with Sketch, templates, image comments, and new API models.
On September 8, 2026, OpenAI officially introduced ChatGPT Images 2.5 with the Sketch feature. It allows users to draw a simple sketch directly in ChatGPT and use it as a visual reference when creating an image.
Sketch is activated with the @Sketch command. Users can define the composition, object placement, or basic shape of a design through a sketch and supplement it with a text prompt.
ChatGPT Images 2.5 introduces visual input
The feature allows a sketch to be used as a visual reference for the generated output. In a brief editorial test by The Verge, the tool turned a simple drawing of a cat into a realistic image based on the provided instructions.
This type of control may be particularly useful for composition drafts, marketing materials, or creatively developing a visual idea. Users can supplement the sketch with a text prompt containing additional requirements for the result.
Templates, comments, and prompt sharing
ChatGPT Images 2.5 adds templates for formats including Poster and Merch. OpenAI has also added comments directly within images.
Another new feature is the ability to share a prompt together with an image.
Rollout is underway for ChatGPT, ChatGPT Work, and Codex users on the web, in desktop apps, and on mobile devices. OpenAI announced availability for users at all levels, but full access may be gradual depending on the region, platform, or plan.
Two new models for the API
For developers, OpenAI has also released two new image models in the API: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. The announcement therefore covers both the ChatGPT interface and applications and services that integrate image generation through the API.
OpenAI claims more natural lighting, better detail preservation, more precise edits, and up to 50% lower latency compared with Images 2.0. Independent comparative benchmarks for quality, speed, and editing reliability are not yet available.
Provenance protection and the risk of realistic forgeries
The new capabilities expand the potential for creating realistic synthetic images. In its system card, OpenAI cites the risk of convincing deepfakes and describes safeguards including input and output filters, C2PA metadata, and the invisible SynthID watermark.
In the near term, it will be important to monitor the actual availability of Sketch and the new API models on individual platforms, as well as independent tests against Images 2.0 and competing models. The practical effectiveness of the declared safeguards against misuse will also be relevant.
Sources
- OpenAI – Introducing ChatGPT Images 2.5 – Confirms the announcement, Sketch, templates and comments, availability in ChatGPT and Codex, and the new API models.
- OpenAI Deployment Safety Hub – ChatGPT Images 2.5 System Card – Documents the stated safeguards, the risk of realistic deepfakes, internal safety evaluation results, and image provenance tools.
- The Verge – ChatGPT Sketch turns your bad drawings into detailed AI images – Independently confirms Sketch’s basic functionality through a brief editorial test.
Verified and updated: 09/09/2026 06:23
