ChatGPT Images 2.5 Speeds Up Image Generation

Key Takeaways

1Images 2.5 is available for all ChatGPT, Work, and Codex users, as well as via the API with two models, Flare and Sunburst

2Image generation latency is reduced by up to 50% compared to the previous version

3New tools such as Sketch, ready-to-use templates, annotation, and prompt sharing are offered

4Controls, C2PA metadata, and invisible watermarking accompany image generation

Why it matters : Images 2.5 aims to enhance the quality, speed, and security of image creation for professional and creative uses.

The new version of Images 2.5 is available to users of ChatGPT, Work, and Codex on desktop, mobile, and web, as well as via the API with two models. The updates include reduced latency, more precise editing, better fidelity to references, and creative tools like Sketch and ready-to-use templates.

General Availability and Two Models in the API

Images 2.5 is accessible to users of ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web. The rollout begins today across all subscription levels. On the API side, two models are launched: GPT-Image-2.5 Flare, the default choice for most applications with the latest improvements, and GPT-Image-2.5 Sunburst, designed for premium visual workflows requiring stricter control over modifications. Both models are available for existing integrations.

Active Protections, C2PA Metadata, and Invisible Watermark

ChatGPT Images 2.5 relies on controls applied to prompts and images to help prevent harmful outcomes. The generated images can be identified through the use of C2PA metadata and an invisible watermark. The publisher states that image generation is designed to be useful, creative, and secure.

Performance: Up to 50% Less Latency

Image generation sees latency reduced by up to 50% compared to version Images 2.0. In the API, GPT-Image-2.5 Flare also announces 50% less latency compared to GPT-Image-2, while maintaining quality and editing gains.

Enhanced Fidelity to References and Visual Understanding

The model performs better with reference photos to transpose subjects into new contexts, styles, and compositions. Subjects appear more recognizable, with more natural lighting and textures, and distinctive features better preserved. Images 2.5 understands complex visual instructions better and manages more elaborate layouts, including transparent backgrounds, while reflecting a visual style more accurately. For teams and businesses, these advancements are presented as making reference-based workflows and complex creative briefs more reliable, with better preservation of visual direction, composition, and details.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Simplified Modification of Specific Elements

The targeted editing process is simplified to modify only the requested element, even in complex scenes. Through the API, it is indicated that one can update a product, background, or text without altering the subject, composition, or branding treatment. During long conversations, the model follows successive instructions more faithfully, retaining previous changes without quality degradation over multiple edits. This consistency is highlighted as useful for applying targeted modifications without needing to rebuild a complete asset.

New Creative Tools: Sketch, Templates, Annotations, and Prompt Sharing

New features provide more control over creation. Sketch allows for a sketch directly in ChatGPT to serve as a visual guide and transform, for example, the layout of a room, the outline of an outfit, or a humorous drawing into a final image, with the option to add style and details. Ready-to-use templates for popular formats like posters or merchandise help structure prompts and quickly customize outputs. Comments can be added to images to guide retouching, and it is possible to share an image with the associated prompt so that others can create their own variant. The publisher emphasizes gains in sharpness, better preservation of reference subjects, and an expanded launch, with a reported weekly volume of over 3 billion generated images.