OpenAI’s Image Generation Gets Sharper with Version 2.5
OpenAI has released version 2.5 of its image generation models, marking a significant improvement in instruction following and visual fidelity. The new models have already been used to generate over 3 billion images across platforms like ChatGPT Images and the GPT-Image API.
The key upgrades include:
- Improved Instruction Following: Models now better understand complex prompts across multiple turns of conversation
- Faster Response Times: Image generation is noticeably quicker than previous versions
- Enhanced Subject Preservation: Reference photos maintain greater consistency in generated images
Two New Model Options
The API now offers two distinct models:
gpt-image-2.5-sunburst: Prioritizes editing precision and complex compositionsgpt-image-2.5-flare: Optimized for speed in everyday image generation tasks
OpenAI recommends Sunburst for workflows requiring detailed edits, while Flare excels at producing high-quality images quickly.
Practical Applications
The updated models support passing multiple reference images—allowing users to create variations or extend existing visuals. For example, you can now provide a photo and ask the model to generate similar scenes with different lighting or compositions.
This release builds on OpenAI’s commitment to improving generative AI tools for both creative professionals and everyday users.