ChatGPT Images 2.5: Everything You Need to Know About OpenAI's New Image Generator (Complete Guide 2026)
OpenAI dropped a major update on September 8, 2026: ChatGPT Images 2.5, the company's new state-of-the-art image generation model. It ships with sharper details, faster generation, a brand-new Sketch feature, and two companion API models — GPT-Image-2.5 Flare and Sunburst.
We've spent the last 48 hours digging through the official system card, testing the new features, and comparing it against every generation of OpenAI's image models to give you the complete picture. Here's everything you need to know.
What Is ChatGPT Images 2.5?
ChatGPT Images 2.5 is OpenAI's latest image generation model, rolled out to all ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web. It replaces GPT Image 2 (April 2026) as the default image engine in ChatGPT, bringing improved image quality, up to 50% faster generation, more precise multi-turn editing, and a suite of new creation tools including the hand-drawn Sketch input and Templates.
For developers, OpenAI also released two new API models — GPT-Image-2.5 Flare (the fast default) and GPT-Image-2.5 Sunburst (precision-oriented) — both accessible via the Images API.
OpenAI reports that users now create more than 3 billion images each week across ChatGPT Images and the GPT-Image API models.
The Evolution: From DALL-E to GPT Image 2.5
Understanding where Images 2.5 sits in OpenAI's image model timeline helps put its improvements in context:
| Year | Model | Milestone |
|---|---|---|
| 2021 | DALL-E | First public AI image generator from OpenAI |
| 2022 | DALL-E 2 | Higher resolution, better quality, still strict filters |
| 2023 | DALL-E 3 | Integrated into ChatGPT, best-in-class prompt following at launch |
| 2025 | GPT Image (gpt-image-1) | DALL-E replaced with language-model-native image generation |
| 2025 Dec | GPT Image 1.5 | Interim update with improved text rendering |
| 2026 Apr | GPT Image 2 / ChatGPT Images 2.0 | O-series reasoning, native 2K output, near-perfect text |
| 2026 Sep | ChatGPT Images 2.5 | Sketch, Templates, 50% faster, Flare + Sunburst API |
The jump from DALL-E 3 to GPT Image was architectural — moving from a separate model bolted onto ChatGPT to a natively integrated image generator. Images 2.5 is the third generation of that architecture, and the first to ship with dedicated creation tools.
Key Features of ChatGPT Images 2.5
We tested each of the new features to see how they perform in real-world use.
1. Enhanced Image Quality — Sharper Details, Natural Lighting, Richer Textures
Images 2.5 produces noticeably better lighting and texture than GPT Image 2. We found it handles skin tones under mixed lighting far more naturally, and fine details like fabric texture and foliage render with less of the "plastic" look earlier models suffered from. Subject preservation from reference photos is also improved — the model keeps key features intact even when you change background or composition.
2. Faster Generation — Up to 50% Lower Latency
OpenAI claims latency is cut by up to 50% compared to Images 2.0. In our testing, first-image generation feels roughly twice as fast — useful when you're iterating on concepts quickly. This is especially noticeable in multi-turn editing sessions where each edit previously required a full regeneration cycle.
3. Precise Multi-Turn Editing
One of the biggest improvements: editing instructions are now followed reliably across multiple turns. We tested a four-edit chain — changing background color, adding an object, adjusting lighting, and cropping — and the model maintained edit consistency without quality degradation. Earlier models had a tendency to "forget" previous edits or introduce artifacts after two or three changes.
4. Sketch Feature (@Sketch)
The new Sketch feature, invoked by typing `@Sketch` in a chat, opens a drawing canvas inside ChatGPT. You can draw a rough visual reference — stick figures, basic shapes, layout guides — and the model uses it alongside your text prompt to generate the final image. We found it works best for compositional guidance: placement of elements, rough proportions, and scene layout. It won't turn your stick figure into a photorealistic portrait (that's not what it's for), but it gives you a level of visual control that text-only prompts can't match.
5. Templates (Poster, Merch, Flyer, Product Photo)
OpenAI introduced templates for common formats, with named options including "Poster" and "Merch." These provide pre-configured dimensions and layout starting points. We tested the Poster template and found it generates properly sized output with sensible margin defaults — saves a round of "crop this to 2:3" editing.
6. Comment Editing
You can now place text comments directly on generated images to specify areas for focused editing. Instead of describing "the thing in the top-right corner," you drop a comment pin and describe the change. This makes complex multi-element edits far more precise.
7. Prompt Sharing
Images 2.5 lets you share the exact prompt behind any image you generate. Other users can take your prompt, swap in their own photos and details, and run the same idea. This is a useful feature for collaborative workflows and inspiration sharing.
Technical Specifications
| Specification | Details |
|---|---|
| Max output resolution | Up to 2K (GPT Image 2 baseline); optional 4K upscaling |
| Aspect ratios | 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 21:9, and more |
| Max images per request | Up to 10 (via API) |
| Reference images per edit | Up to 16 |
| Context window | 400K tokens (Sunburst/Flare API) |
| Latency reduction | Up to 50% vs GPT Image 2 |
| Safety | C2PA metadata, invisible watermarking, prompt/output classifiers |
| Availability | All ChatGPT tiers, ChatGPT Work, Codex (desktop, mobile, web) |
API Models: GPT-Image-2.5 Flare vs Sunburst
OpenAI released two API models alongside ChatGPT Images 2.5. We compared both.
| Feature | GPT-Image-2.5 Flare | GPT-Image-2.5 Sunburst |
|---|---|---|
| Positioning | Default for most apps | Premium precision workflows |
| Latency vs GPT Image 2 | 50% lower | Similar to GPT Image 2 |
| Editing precision | High | Highest — tighter control across edits |
| Generation speed | Fastest | Longer generation times |
| Best for | High-volume content, creator content, rapid prototyping | Production-ready campaigns, polished product imagery |
| Price (Image input) | $8.00 / 1M tokens | $8.00 / 1M tokens |
| Price (Cached input) | $2.00 / 1M tokens | $2.00 / 1M tokens |
| Price (Image output) | $30.00 / 1M tokens | $30.00 / 1M tokens |
| Price (Text input) | $5.00 / 1M tokens | $5.00 / 1M tokens |
Both models share the same pricing. The difference is in latency and editing precision — Flare is faster, Sunburst is more deliberate. For most use cases, Flare is the right choice. Sunburst shines when you need pixel-level control over a production asset.
Images 2.5 vs GPT Image 2 vs DALL-E 3: Comparison Table
We compared all three across the dimensions that matter for creators:
| Feature | DALL-E 3 | GPT Image 2 | **ChatGPT Images 2.5** |
|---|---|---|---|
| Image Quality | ★★★ | ★★★★ | ★★★★★ |
| Generation Speed | Slow | Fast | Fastest (50% faster) |
| Text Rendering | ⚠️ | ✅ | ✅ (best-in-class) |
| Multi-Turn Editing | ❌ | ⚠️ | ✅ Reliable |
| Sketch Feature | ❌ | ❌ | ✅ |
| Templates | ❌ | ❌ | ✅ |
| API Models | DALL-E 3 API (retired) | gpt-image-2 | Flare + Sunburst |
| Max Resolution | 1024x1024 | 2K native | 2K (4K optional) |
| C2PA Metadata | ❌ | ✅ | ✅ |
| NSFW Content | ❌ Blocked | ❌ Blocked | ❌ Blocked |
Pricing
ChatGPT Images 2.5 is available to all ChatGPT users as part of their existing subscription:
- Free tier: Limited generations per day
- ChatGPT Plus: $20/month — full access
- ChatGPT Work: Included in Work subscription
- Codex: Included in Codex subscription
For API users, both GPT-Image-2.5 Flare and Sunburst share this pricing:
| Cost Component | Price per 1M tokens |
|---|---|
| Image input tokens | $8.00 ($2.00 cached) |
| Image output tokens | $30.00 |
| Text input tokens | $5.00 ($1.25 cached) |
| Text output | Not billed (models output images) |
What Images 2.5 Cannot Do (Content Restrictions)
Like every OpenAI image model before it, ChatGPT Images 2.5 blocks all NSFW content — including nudity, sexually explicit imagery, suggestive poses, adult scenarios, violent content, and hateful imagery. The filters are as strict as ever, and the Sketch feature and comment editing don't bypass them.
We tested this directly across multiple prompt variations. The model consistently refused to generate any adult-oriented content, even with careful prompt engineering. OpenAI's safety classifiers remain a hard ceiling.
This means if your work involves adult content, uncensored creative expression, or NSFW art, ChatGPT Images 2.5 simply isn't the right tool.
The alternative: If you need a truly uncensored AI image generator, platforms like HackAIGC offer native uncensored image generation — no jailbreaks, no workarounds, no hidden filters.
FAQ
What is ChatGPT Images 2.5?
ChatGPT Images 2.5 is OpenAI's latest image generation model, released September 8, 2026. It succeeds GPT Image 2 and introduces Sketch, Templates, 50% faster generation, and better multi-turn editing.
How is Images 2.5 different from DALL-E?
DALL-E was OpenAI's original image model family. GPT Image (released 2025) replaced DALL-E with a language-model-native architecture. Images 2.5 is the third generation of that architecture — far better quality, faster, and with creation tools DALL-E never had.
How do I use the Sketch feature?
Type `@Sketch` in a ChatGPT chat to open the drawing canvas. Draw your visual reference, then write your text prompt alongside it. The model combines both inputs to generate the final image.
Is Images 2.5 available in the API?
Yes, via two models: GPT-Image-2.5 Flare (fast, high quality) and GPT-Image-2.5 Sunburst (precision editing, longer generation). Both are available through OpenAI's Images API.
Does ChatGPT Images 2.5 support NSFW content?
No. Like every OpenAI image model, Images 2.5 blocks all NSFW/adult content. For uncensored image generation, you'll need a dedicated platform.
What are GPT-Image-2.5 Flare and Sunburst?
They are the two API variants of Images 2.5. Flare is the default for most applications — it offers the same quality as Images 2.5 at 50% lower latency than GPT Image 2. Sunburst is for premium workflows requiring tighter editing control, with longer generation times.
Related Articles
- DALL-E Retired: Best Uncensored Alternatives 2026
- How to Generate NSFW AI Images: Complete Guide 2026
- HackAIGC Image Generator: Full Feature Guide 2026
Ready to create without limits? Try HackAIGC — the truly uncensored AI platform for chat, image, and video generation.
Generate uncensored images: HackAIGC Uncensored Image Generator
Generate uncensored videos: HackAIGC Uncensored Video Generator