GPT Image 2.5 is an advanced image generation and editing model launched by OpenAI. It is designed to deliver sharper image quality, higher fidelity to reference images, and more refined image optimization. The model can generate more natural lighting and textures, preserve subject features more reliably, and execute complex visual instructions with greater precision in multi-round editing scenarios.
GPT Image 2.5 Sunburst is the precision tier of OpenAI's GPT Image 2.5 image generation and editing family. It builds on Flare with tighter editing accuracy and multi-round controllability — edits change only the designated area without drifting — suited to production-ready campaign key visuals and polished product imagery.
GPT Image 2.5 Flare is the speed-focused tier of OpenAI's GPT Image 2.5 image generation and editing family. It cuts generation latency by up to 50% versus GPT Image 2 while sharpening detail, lighting and texture, making it the default choice for high-volume generation, creator content and rapid prototyping.
Fast on Flare, precise on Sunburst
Two tiers of OpenAI's latest image generation: Flare for high-volume speed (~50% lower latency than GPT Image 2), Sunburst for surgical multi-round editing — both with exclusive xhigh/max quality tiers.
Prompt
Keep the portrait, blue backdrop and printed-photo edges exactly the same; change only the outfit to a tailored charcoal suit with subtle fabric texture...
Flare is the default — fast and better than 2.0. Sunburst trades some speed for surgical editing precision.
The everyday variant: high-volume generation, creator content and rapid prototyping. Higher quality than 2.0 with latency cut by ~50% — sharper detail, more natural light, richer texture.
Builds on Flare with tighter editing accuracy and multi-round controllability — changes apply only to the designated area and multi-round edits never drift.
Both variants accept the exclusive xhigh / max quality tiers and share the same request shape as GPT Image 2 — pick a tier by swapping the model ID.
The latest GPT Image generation ships in two tiers — Flare for speed, Sunburst for precision — sharing one async workflow and six quality levels.
Generation latency drops by up to 50% compared with GPT Image 2, so drafts and iterations come back sooner.
Reference-image edits hold on to subjects, shapes and identity with noticeably better fidelity than the previous generation.
Lighting falls more naturally and material textures — ceramic, fabric, metal, skin — render with crisper micro-detail.
Sequential edits keep labels, edges, geometry and approved details intact across rounds instead of drifting with every change.
Six quality tiers including exclusive xhigh/max fidelity, transparent backgrounds, and up to 16 reference images — on one asynchronous endpoint.
auto, low, medium, high plus two 2.5-exclusive tiers — xhigh and max for the hardest images.
Sixteen aspect ratios across the 1K/2K/4K pixel tiers, from square social assets to 21:9 cinematic banners.
Text-to-image and reference-image editing through one async task API. Transparent-background PNG output at medium quality and above.
Sample prompts showcasing what GPT Image 2.5 handles best — copy, adapt, and generate.
A square studio product photo of a cobalt-blue ceramic mug on a pale gray backdrop, soft light from the left, handle to the right. Preserve the ceramic texture; leave the upper-left quarter empty for headline copy. No text, no logos.
Change only the background from pale gray to warm beige. Keep the mug's shape, color, handle direction, lighting and shadows exactly as they are; preserve the empty headline space.
An Instagram Story-style shot of a cosplayer crouching and facing the camera, playful hand gestures, dramatic rim lighting, magazine-cover composition with clean negative space on top.
Keep the portrait, blue backdrop and printed-photo edges exactly the same; change only the outfit to a tailored charcoal suit with subtle fabric texture. Photorealistic studio lighting.
One async endpoint for generation and editing. Submit a task, poll the status endpoint, and download your images.
import requests import time API_BASE = "https://api.apipod.ai/v1" headers = {"Authorization": "Bearer $API_KEY"} # 1. Submit generation task resp = requests.post( "https://api.apipod.ai/v1/images/generations", headers=headers, json={ "model": "gpt-image-2.5", "prompt": "A cobalt-blue ceramic mug on a pale gray studio backdrop...", "quality": "xhigh", "aspect_ratio": "1:1", "resolution": "2K" } ) task_id = resp.json()["data"]["task_id"] # 2. Poll status until completed while True: res = requests.get( f"https://api.apipod.ai/v1/images/status/{task_id}", headers=headers ).json() if res["data"]["status"] == "completed": print(res["data"]["result"][0]) break time.sleep(2)
Everything you need to know about GPT Image 2.5 on APIPod.
GPT Image 2.5 is OpenAI's latest image generation and editing family, released on September 8, 2026 across ChatGPT, ChatGPT Work and Codex plans. It ships in two tiers — Flare (speed) and Sunburst (precision) — improving on GPT Image 2 with lower latency, stronger subject preservation from reference images, more natural lighting and material texture, and better consistency across continuous edits.
Three things: Flare cuts generation latency by ~50% versus GPT Image 2.0; fidelity improves across the board (sharper detail, more natural light, richer texture, better subject fidelity in reference edits); and both tiers add exclusive xhigh and max quality levels on top of low/medium/high/auto. Every other parameter, endpoint and response field is identical, so migrating is a one-line model ID change.
Flare is the default, recommended tier — fast and better than 2.0, ideal for high-concurrency batch generation, creator content and rapid prototyping. Sunburst is the precision flagship: it builds on Flare with tighter editing accuracy and multi-round controllability, for production-ready campaign key visuals and polished product imagery. Pick a tier by using the public model ID gpt-image-2.5-flare or gpt-image-2.5-sunburst.
Set background to transparent and quality to medium or higher. The result is an alpha-channel PNG, ready for product listings, stickers and layered compositions without manual cutout.
Yes. Images generated through APIPod can be used in commercial projects, subject to OpenAI's usage policies and applicable content guidelines.
Billing is per image, based on the resolution tier (1K/2K/4K) and quality you request. Failed provider-side generations follow the channel's billing rules. Check the Pricing page for the current per-image rates.
Flare for speed, Sunburst for precision — one API key, one async endpoint.
