
Meet GPT-Image-2.5 Flare and Sunburst. Introducing new image models in the API, with sharper detail, stronger style adherence, and more control over edits. https://t.co/dirY4ANhpr
The Hook
“Today we're releasing two new image models in the API: GPT-Image-2.5 Sunburst and Flare.”
Product Reveal
“Sunburst gives you sharper detail, stronger style adherence, and far better control over edits across multiple turns.”
Feature Teaser
“It can follow complex, highly specific spatial instructions that used to trip up earlier models.”
Feature Teaser
“Flare is built for speed. It's more than three times faster than GPT-Image-2 while retaining high quality.”
Feature Teaser
“Both models now support native transparency, preserving semi-transparent elements like wings, glass, and shadows.”
Call to Action
“Both GPT-Image-2.5 Sunburst and Flare are available today in the OpenAI API. We can't wait to see what you build.”
Synthesizable in Remotion · key components of the presenter-technical-benchmark-split template.
A central vertical divider with a circular double-ended arrow drag handle moves horizontally across two overlaid images, clipping GPT-Image-2 on the left and Sunburst on the right.
Use useCurrentFrame() with interpolate() or spring() to calculate divider position from 0 to 100%. Render container with overflow hidden. Render the left image inside a wrapper with clipPath: `inset(0 ${100 - dividerX}% 0 0)` and right image with `inset(0 0 0 ${dividerX}%)`. Position a centered pill handle at left: `${dividerX}%` with transform: 'translateX(-50%)'.Two rounded rectangle viewport cards displaying real-time running second counters at bottom left that halt and switch status from Waiting to Complete when generation finishes.
Compute elapsed seconds via `(frame / fps) * speedMultiplier`. Conditionally cap the counter at model completion timestamps (14.3s for Flare, 47.8s for GPT-Image-2). Render numeric text formatted as `elapsed.toFixed(1) + 's'` and transition badge text from light grey Waiting to green Complete with an opacity crossfade.
Two transparent PNG cutouts rendered side by side over an animated background fill that snaps between #191919 dark and #DE3896 bright magenta, demonstrating alpha blending quality.
Wrap image elements in an AbsoluteFill whose backgroundColor is driven by frame index using interpolateColors(). Place PNG assets with mixBlendMode: 'normal' and add scale interpolation to zoom into wing membrane details on keyframe triggers.
A horizontal bottom-anchored sequence ribbon displaying numbered prompt step cards with an active green selection pill outlining the currently previewed frame.
Render an array of step thumbnails in a flex container with overflow-x auto. Map active index to current frame intervals. Apply border: active ? '1.5px solid #0E7F5E' : '1px solid rgba(0,0,0,0.1)' and scale up active thumbnail slightly using spring().
Purpose. Establish institutional credibility and present the core premise of two specialized image models.
Execution. Medium studio lock-off with crisp typography introducing GPT-Image-2.5 Sunburst & Flare over the engineer.
Purpose. Demonstrate surgical consistency across successive prompt edits with an interactive before/after split slider.
Execution. Ivory canvas hosting a side-by-side interactive split slider tracking sequential edits from jacket recolor to watercolor transformation.
Purpose. Prove spatial reasoning and fine-grained prompt compliance that previous generative models failed.
Execution. Curated 5-up product showcase displaying intricate analog clock hands, floor-specific architecture, and facial micro-features.
Purpose. Quantify Flare's dramatic latency advantage over previous generation models.
Execution. Head-to-head live timer playback at 4x speed showing Flare completing in 14.3 seconds while GPT-Image-2 lags beyond 47 seconds.
Purpose. Showcase complex edge fidelity and partial alpha transparency on intricate subjects.
Execution. High-magnification dragonfly wing inspection dynamically toggled against pitch black and hot magenta backgrounds to reveal clean alpha margins.
Purpose. Guide developers to API access and wrap with corporate identity mark.
Execution. Return to presenter in studio delivering concluding remarks with dynamic brand geometry.