Introduction

OpenAI just turned the rumor mill into a release note. On September 8, 2026, the company officially introduced ChatGPT Images 2.5 — a new state-of-the-art image model for ChatGPT, plus two API siblings (GPT‑Image‑2.5 Flare and GPT‑Image‑2.5 Sunburst) aimed at production creative workflows.

If you make stills for moodboards, merch, campaigns, or multi-turn edit sessions, this is the drop that finally pairs sharper fidelity with up to 50% lower generation latency versus Images 2.0 — and it ships with product tools that feel built for how artists actually iterate.

What’s New in Images 2.5

OpenAI says people already create more than 3 billion images a week across ChatGPT Images and the GPT‑Image API models. Images 2.5 leans into that volume with a few creator-facing upgrades:

  • Sharper fidelity from references — better likeness preservation for people, pets, and product shots, with more natural lighting and richer textures when you transform a subject into new settings or styles.
  • Precision editing — change one element (copy, background, prop) while keeping composition and brand treatment intact.
  • Multi-turn consistency — longer ChatGPT conversations stay steadier; earlier edits are less likely to melt as you refine.
  • Smarter layouts — stronger handling of complex visual instructions, real-world information in graphics, and even transparent backgrounds.

Sketch, templates, and shareable prompts

The model launch comes with ChatGPT product features that matter as much as the checkpoint:

  • Sketch — draw directly in ChatGPT (@Sketch) and use the doodle as a layout/reference for the final image. Room plans, outfit silhouettes, rough storyboards — then describe the style you want.
  • Templates — starter formats like posters and merch so you’re not staring at a blank prompt box.
  • Inline comments on images — mark what to fix without rewriting the whole brief.
  • Share prompts with images — hand someone the recipe so they can remix it with their own photos.

Flare vs Sunburst in the API

For builders, OpenAI split the API into two flavors:

  • GPT‑Image‑2.5 Flare — the default path for most apps: higher quality than GPT‑Image‑2 at roughly half the latency. Built for social/creator content, product experiences, visual search, prototyping, and high-volume generation.
  • GPT‑Image‑2.5 Sunburst — slower, tighter control for premium campaign creative and polished product imagery where every edit pass has to land.

Early quotes on OpenAI’s post from teams like Higgsfield AI, Adobe, Manus, and Runway lean hardest on a shared theme: the model is better at understanding what not to change — the character, composition, and visual identity that make iterative creative work possible.

Pro Tip

Treat Flare as your daily driver for volume and exploration, then promote hero frames to Sunburst (or a careful multi-turn ChatGPT edit stack) only when brand lock, typography, and subject fidelity have to survive a dozen revisions. Pair Sketch when layout is the hard part — draw the hierarchy first, then let the model fill the light.

Availability and Safety Notes

Images 2.5 is rolling out to ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. Flare and Sunburst are live in the Images API (check OpenAI’s pricing page for current rates). Safety continues to use prompt/image checks, C2PA metadata, and invisible watermarking; OpenAI points to a system card for the evaluation details.

Conclusion

ChatGPT Images 2.5 is less a flashy “new filter” and more a workflow release: faster generation, stickier multi-turn edits, Sketch-as-briefing, and an API split that finally admits speed and precision are different jobs. For prompt engineers and digital artists, the sunny takeaway is simple — iterate more, melt less, and keep your subject looking like themselves while the world around them blooms.

—Aurelia ♡