GPT Image 2.5 brings sharper detail, element-level editing, faithful reference photos, and up to 50% lower latency across two API models. While GPT Image 2.5 rolls out, create with the GPT Image workflow on Seedance.io. Try GPT Image 2.5 online for free!
Model overview
GPT Image 2.5 — announced as ChatGPT Images 2.5 on September 8, 2026 — is OpenAI's new state-of-the-art image model. It improves lighting, texture, and text rendering, preserves the subjects in your reference photos, and keeps edits consistent across multiple turns, at up to 50% lower latency than Images 2.0.
The API splits GPT Image 2.5 in two. gpt-image-2.5-flare is OpenAI's fastest model for high-quality everyday generation — creator content, product visuals, and high-volume pipelines. gpt-image-2.5-sunburst is the most capable option for image generation and editing, built for workflows where editing precision matters most. Both share identical token pricing, support inpainting, and carry the 2026-09-08 snapshot.
What is new
The GPT Image 2.5 release targets the places where image generation breaks down in real production: fidelity, targeted edits, subject consistency, and speed.
GPT Image 2.5 renders richer textures, more natural light, and more accurate real-world information. It handles complex layouts — including transparent backgrounds — and holds a visual style more consistently across a set of related images.
Ask for one change — a product, a background, a line of copy — and GPT Image 2.5 updates that element while preserving the composition, subjects, and brand treatment around it. Earlier edits carry through long conversations without quality degradation.
Attach one or more reference images and GPT Image 2.5 preserves the character, composition, and visual identity of your subject through generation and later edits — the biggest single upgrade for character work and product imagery.
In ChatGPT, type @Sketch to draw a visual reference directly, start from templates such as Poster and Merch, place comments on an image to direct a focused edit, and share prompts so others can rerun them with their own photos.
Production control
A usable GPT Image 2.5 image survives feedback rounds. GPT Image 2.5 is tuned for exactly that: repeated, targeted revisions that protect the decisions already approved.
01
GPT Image 2.5 instruction following holds up across long editing conversations, so a campaign visual can go through round after round of copy, layout, and detail changes without the rest of the image drifting.
02
Flare answers fastest and suits everyday, high-volume generation; Sunburst trades speed for the tightest control across edits. Both cost the same per token, so the choice is about workflow, not budget.
03
GPT Image 2.5 images carry C2PA content-credentials metadata, and OpenAI is adding Google DeepMind's SynthID invisible watermarking across ChatGPT, Codex, and the API, so output stays identifiable after routine edits.
Creative workflows
GPT Image 2.5 is aimed at imagery that must stay on-brand through edits, ship at volume, and increasingly end up in motion.
01
Product visuals that survive revisions
Generate polished GPT Image 2.5 product scenes, then swap backgrounds, seasons, or copy without disturbing the composition the team already approved.
02
High-volume social content with Flare
Flare targets creator and social pipelines — daily variants, thumbnails, and channel-specific layouts at 50% lower latency than Images 2.0.
03
Campaign creative with Sunburst
Sunburst is built for production-ready campaign visuals and polished product imagery where every revision must change one element and preserve the rest.
04
Turn stills into video on Seedance.io
Use a finished GPT Image 2.5 image as a start frame, style reference, or storyboard panel, then animate it with Seedance 2.5, MiniMax H3, and other video models in the same workspace.
Four-step workflow
GPT Image 2.5 handles composition and identity; the video models handle motion. Seedance.io keeps both steps in one workspace, with the GPT Image workflow available today.
Sign in on Seedance.io and open the image workspace. Describe the subject, setting, composition, lighting, and any required text, or add a reference image that should guide the result.
Pick the GPT Image 2.5 size and quality for the intended channel, generate, and judge the result against the brief — composition, message, hierarchy, and detail, not style alone.
Tell GPT Image 2.5 what should change and what must stay. Replace a background, correct a line of copy, or adjust the light one instruction at a time — the workflow GPT Image 2.5 is optimized for.
Keep the GPT Image 2.5 image as a standalone asset, or carry it into image-to-video and Canvas to build motion around the visual direction you approved.
Visual examples
Selected images from OpenAI's launch announcement — a retrofuturist illustration, mid-century poster grids, an ornate wedding invitation, vintage stamps, a tile mosaic, and a cinematic night scene. Judge every GPT Image 2.5 result by clarity, composition, and fit for the final channel.
GPT Image 2.5
GPT Image 2.5
GPT Image 2.5
GPT Image 2.5
GPT Image 2.5
GPT Image 2.5Prompt guide
GPT Image 2.5 rewards structured briefs and short, focused follow-ups. Communicate intent and priorities, then refine one decision at a time.
Prompt 01
Start with the subject and action, then add setting, camera distance, perspective, lighting, mood, and where key elements belong. If copy will sit above the subject, reserve that space in the brief.
Prompt 02
Put required wording in quotation marks and state where it belongs. Describe hierarchy, type treatment, color direction, and audience. Text rendering improved in GPT Image 2.5, but critical copy still deserves a character-by-character check.
Prompt 03
The headline upgrade in GPT Image 2.5 is edit reliability. Instead of one oversized prompt, iterate: name the edit target, the requested change, and the details GPT Image 2.5 must protect — then repeat until approved.
Model comparison
The 2.5 release splits into two models with the same price and different priorities, while GPT Image 2 stays available. Compare the documented positioning before choosing.
01
GPT Image 2.5 Flare is the default: OpenAI's fastest model for high-quality everyday generation, delivering higher quality than GPT Image 2 at 50% lower latency. GPT Image 2.5 Sunburst is the most capable for generation and editing, built for premium workflows where editing precision matters most — at the cost of longer generation times.
02
The GPT Image 2.5 models add sharper detail, more natural lighting, more faithful reference photos, and far steadier multi-turn edits, at the same token rates. GPT Image 2 is not being retired, so pipelines tuned to it can migrate on their own schedule.
03
Both target professional image work. GPT Image 2.5 centers precision editing, reference fidelity, and speed; Google positions Nano Banana Pro for complex graphic design, accurate text, and Search-grounded real-world context. Run the same brief through both before committing.
At-a-glance
These are workflow-level differences, not a quality ranking. Model versions, output options, and availability change; open the linked primary sources before making a production decision.
GPT-Image-2.5 Flare
GPT-Image-2.5 Sunburst
GPT Image 2
Nano Banana Pro
GPT Image 2.5 Flare and Sunburst share identical token pricing: $5 per 1M text input tokens, $8 per 1M image input tokens, and $30 per 1M image output tokens, with cached-input discounts — the same rates as GPT Image 2.
Primary sources
Start with starter credits
GPT Image 2.5 is rolling out across ChatGPT and the OpenAI API. On Seedance.io you can create with the GPT Image workflow today — free starter credits at sign-up — and this page tracks the GPT Image 2.5 models as they reach the workspace.
Start CreatingFAQ
Quick answers about the GPT Image 2.5 release, Flare and Sunburst, pricing, editing, watermarking, and how it fits the Seedance.io workspace.
Connected workflow
A strong still is often the first frame of something bigger. Keep GPT Image 2.5 generation, review, iteration, and motion planning in one workspace so the approved GPT Image 2.5 direction survives the handoff.
01
One workspace for image and video models
Create with the GPT Image workflow, compare against Seedream 5.0 and Nano Banana, and animate the winner with Seedance 2.5 or MiniMax H3 — without moving projects between tools.
02
Protect approved details during edits
Use focused revision instructions to preserve the subject, crop, palette, or composition while changing the one detail that still blocks approval.
03
Carry still images into motion
Move a selected concept into image-to-video and Canvas workflows without losing the starting frame, story beat, or art direction that made it useful.
Keep exploring
Different briefs benefit from different visual engines and follow-up tools. Compare the workflow you need, then choose the shortest path from a GPT Image 2.5 idea to a usable asset.
Practical guides
Use these guides to strengthen GPT Image 2.5 prompts, turn still images into motion, and compare the wider generation workflow before production.
Related models
Compare GPT Image 2.5 against the other image models on Seedance.io, or continue from a finished still into video generation.
GPT Image 2.5 · Released Sep 8, 2026
Start with a GPT Image 2.5 prompt or reference image, refine one decision at a time, and animate the strongest result into video.
Start Now