A faster, sharper ChatGPT image model
gpt-image-1.5 builds on the native multimodal architecture of its predecessor — text and images processed together in a unified neural network — but delivers the output 4× faster at 20% lower cost. The most visible improvements are in editing precision and small text rendering. Edits now preserve lighting, composition, likeness, and appearance while applying changes exactly as instructed. All the world-knowledge grounding from GPT-4o carries over: accurate depictions of logos, flags, landmarks, scientific diagrams, and knowledge-rich infographics.Capabilities
Precise editing
More reliable instruction following during edits — preserves lighting, composition, subject likeness, and overall appearance while applying exactly the specified change.
World-knowledge grounding
GPT-4o’s knowledge enables accurate rendering of real-world entities: brand logos, national flags, scientific diagrams, famous landmarks, and contextual event imagery.
Up to 10 references
Accepts up to 10 reference images in a single generation — combine products, people, backgrounds, and styles for complex composite outputs.
Specifications
| Feature | Details |
|---|---|
| Model ID | gpt-image-1.5 |
| Speed vs. gpt-image-1 | 4× faster |
| Cost vs. gpt-image-1 | 20% cheaper |
| Reference images | Up to 10 |
| Architecture | Native multimodal (unified text + image) |
| Released | December 16, 2025 |
How to use
Write your prompt
Include real-world references, specific knowledge, or detailed text requirements in your prompt. gpt-image-1.5 excels when your prompt requires world knowledge.
Attach reference images (optional)
Upload up to 10 reference images for compositing or style anchoring.
Prompting tips
- Name real-world subjects directly — “A poster celebrating the Apollo 11 mission, with an accurate depiction of the lunar module” leverages GPT-4o’s world knowledge for accuracy.
- Use it for infographic copy — Dense, labeled infographics with specific text content render cleanly. Describe exact headings, labels, and callout text you want.
- For editing, be specific about what stays — “Change the background to a forest but keep the subject’s lighting, pose, and outfit exactly as-is.”
Example prompts
A detailed infographic explaining photosynthesis with labeled diagrams, clean white background, educational illustration style, accurate scientific labels.
A World War II era propaganda poster aesthetic with bold typography reading “INNOVATION NEVER SLEEPS”, dark navy and gold, vintage texture.
A modern smartphone mockup showing a fitness app dashboard with charts, clean UI design, soft shadow.
Compare models
| Model | Speed | Text rendering | World knowledge | Best for |
|---|---|---|---|---|
| ChatGPT Image 1.5 | Fast (4× v1) | Superior (dense + small) | Excellent | Infographics, fast knowledge-grounded |
| ChatGPT Image | Up to 2 min | Best-in-class | Excellent | Complex knowledge-grounded, 10 refs |
| Ideogram v3 | Fast | ~90–95% | Limited | Typography, brand design |
ChatGPT Image 1.5 is the updated version of gpt-image-1. If you need the absolute maximum in knowledge-grounded accuracy and can wait for longer generation times, the original ChatGPT Image is still available.