ChatGPT Images 2.5 Adds Two API Tiers and Sketch Input

My notes

Summary

OpenAI released Images 2.5, a new image model for ChatGPT with sharper detail, more reliable multi-turn editing, and up to 50% faster generation than Images 2.0. Two matching API models launched alongside it, GPT-Image-2.5 Flare as the fast and cheap default and GPT-Image-2.5 Sunburst as the higher-precision, slower option, plus new ChatGPT features: sketch-to-image, format templates, and shareable prompts.

Key Insight

• Scale context: OpenAI says over 3 billion images per week are already generated across ChatGPT Images and the GPT-Image API, so image generation has moved well past novelty into production-volume usage.
• Two-tier API pattern: Flare is the default for high-volume and low-latency use such as social content, product photos, and rapid prototyping, while Sunburst trades speed for precision on production-grade creative like campaign assets and polished product imagery. That is a useful cost and quality lever for anyone building on the API rather than just using ChatGPT directly.
• Multi-turn edit consistency improved: earlier edits in a conversation now persist without degrading as new edits stack on top, previously a common failure mode where quality drifted over iterative editing.
• Reference-photo fidelity improved: better preservation of subject identity and distinctive features when placing a real photo into new settings, styles, or compositions, relevant for personalized or branded creative that needs the same product or model across multiple scenes.
• Handles transparent backgrounds and complex layouts better, which is directly useful for ecommerce product shots and marketing asset pipelines that need clean cutouts.
• New sketch feature: draw a rough layout or shape directly in ChatGPT and the model builds a finished image from it, a lower-friction way to control composition precisely without a full text prompt.
• Provenance: continues C2PA metadata and invisible watermarking on generated images, relevant if publishing AI images where authenticity or detection matters.