OpenAI Unveils ChatGPT Images 2.5: Sketch Feature Turns Doodles Into Professional AI-Generated Art

OpenAI Releases ChatGPT Images 2.5 With Major Speed and Quality Improvements

OpenAI announced ChatGPT Images 2.5 on Tuesday, introducing substantial upgrades to its AI image generation capabilities that serve over 3 billion images created weekly across ChatGPT Images and GPT-Image-2.5 models4

4

. The update delivers up to 50% faster image generation compared to Images 2.0, while producing images with natural lighting and richer textures1

1

4

4

. The enhanced model demonstrates improved ability to follow editing instructions across multiple turns and better preserves subjects in reference photos while maintaining distinctive features across different settings and visual styles3

3

4

4

.

ChatGPT Sketch Feature Transforms Rough Doodles Into Detailed Images

The standout addition is the ChatGPT Sketch feature, which allows users to draw directly inside ChatGPT and transform basic sketches into polished, detailed images1

1

2

2

. Users activate the feature by typing @Sketch into the chat box, opening a drawing window for creating rough layouts, outfit concepts, or simple doodles that AI refines into complete images1

1

2

2

. Early testing confirmed the tool successfully converts mouse-drawn sketches into realistic photos based on user instructions1

1

. This sketch-based editing approach addresses a common challenge in AI-driven creative tools: communicating precise positioning and layout requirements without relying solely on text descriptions2

2

.

Enhanced Editing Tools Enable Precise Visual Modifications

ChatGPT Images 2.5 introduces enhanced editing tools that allow users to leave comments directly on specific parts of images requiring changes, eliminating the need for purely text-based modification requests1

1

3

3

. Testing demonstrated the system's ability to make targeted adjustments, such as changing eye color from blue to green, while preserving other image elements1

1

. The model excels at focused edits while keeping other details unchanged, even with complex subjects and backgrounds, and maintains consistency across multiple edits during extended conversations without quality degradation4

4

. Users can now type and place text directly, sketch objects to be added, and make precise changes through visual indicators rather than verbal descriptions alone3

3

.

Improved Performance With People, Pets, and Reference Photos

The updated image engine demonstrates significant advancement in preserving the likeness of people and pets, an area where generative AI systems have historically struggled3

3

. Testing revealed the model's ability to capture distinctive features and expressions, with one example showing how it preserved a cat's "wonderfully unimpressed expression" while transforming the photo into various artistic styles including film noir detective, stained-glass window, and Renaissance portrait3

3

. The system better handles reference photos, allowing image subjects to appear more recognizable while maintaining their unique characteristics across different settings and visual styles4

4

. This improvement extends to practical applications like tattoo design, where the tool successfully superimposed new designs next to existing tattoos and incorporated specific user feedback through visual editing3

3

.

Developer Access Through New API Models

OpenAI released two new models in the API for developers seeking to integrate these capabilities4

4

. GPT-Image-2.5 Flare delivers the same quality, editing, and speed improvements available to ChatGPT users, while GPT-Image-2.5 Sunburst offers additional precision for detailed creative work with longer generation times4

4

. Both API models improve handling of complex visual instructions and demonstrate more accurate processing of real-world information, including better management of complex layouts with transparent backgrounds and enhanced reflection of visual styles4

4

.

Addressing Creative Expression While Acknowledging Ethical Concerns

Adele Li, OpenAI's product lead on images, emphasized that tools like Sketch aim to make users more engaged in creative expression rather than passive recipients of AI output3

3

. The goal is shifting away from a singular AI aesthetic toward enabling individual creative expression, with Li stating, "I don't want to be able to see ChatGPT in the world. I want people to be able to generate and express their own individualism"3

3

. However, the release acknowledges ongoing ethical implications, as generative AI systems were trained on work from real artists, typically without consent or compensation3

3

. OpenAI maintains existing safeguards with prompt checks on inputs and images to prevent harmful outputs, while continuing to use C2PA metadata and invisible watermarking to identify AI-generated images4

4

. The update is available to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web platforms1

1

4

4

.